detached 689069f

Expand model catalog and improve session resume

Add built-in provider catalog and support for additional provider backends, including Bedrock, Vertex, Azure OpenAI, GitHub Copilot, OpenAI Responses, and extra user-configured providers/models.

Improve TUI session resume by restoring last-turn context usage and tail-rendering long transcripts for faster first paint.

Handle Gemini image models without tools, persist generated images to the working directory, and make Anthropic transcript replay tolerate assistant-side image artifacts.
commit
689069f11a0c71dc3d049dc9e3f547eea4cbdfeb
author
patriceckhart · · 2026-05-24 08:14 UTC
committer
patriceckhart · · 2026-05-24 08:15 UTC
parent
91c696d
signature
unsigned
29 files · +3761 -82
Unified
UnifiedSplit
A .gitignore +75 -0
@@ -0,0 +1,75 @@
1+.DS_Store
2+._*
3+Icon
4+.DocumentRevisions-V100
5+.fseventsd
6+.Spotlight-V100
7+.TemporaryItems
8+.Trashes
9+.VolumeIcon.icns
10+.com.apple.timemachine.donotpresent
11+
12+*~
13+.fuse_hidden*
14+.Trash-*
15+.nfs*
16+
17+Thumbs.db
18+ehthumbs.db
19+Desktop.ini
20+$RECYCLE.BIN/
21+
22+.vscode/*
23+!.vscode/settings.json
24+!.vscode/tasks.json
25+!.vscode/launch.json
26+!.vscode/extensions.json
27+.history/
28+*.vsix
29+
30+.idea/
31+*.iml
32+
33+*.swp
34+*.swo
35+\#*\#
36+.\#*
37+
38+*.exe
39+*.exe~
40+*.dll
41+*.so
42+*.dylib
43+*.test
44+*.out
45+coverage.txt
46+profile.out
47+go.work
48+go.work.sum
49+.env
50+.env.local
51+.env.*.local
52+
53+bin/
54+dist/
55+
56+node_modules/
57+npm-debug.log*
58+yarn-debug.log*
59+yarn-error.log*
60+.pnpm-debug.log*
61+
62+__pycache__/
63+*.py[cod]
64+*.egg-info/
65+
66+auth.json
67+sessions/
68+logs/
69+
70+*.bak
71+*.orig
72+*.rej
73+.tmp/
74+tmp/
75+npm/
M internal/agent/build.go +179 -33
@@ -134,6 +134,8 @@ func defaultModelForProvider(prov string) string {
134134 return "gpt-5"
135135 case "openai-codex":
136136 return "gpt-5.5"
137+ case "openai-responses":
138+ return "gpt-5"
137139 case "kimi":
138140 return "kimi-for-coding"
139141 case "deepseek":
@@ -142,11 +144,75 @@ func defaultModelForProvider(prov string) string {
142144 return "gemini-2.5-pro"
143145 case "ollama":
144146 return ""
147+ case "moonshotai", "moonshotai-cn":
148+ return "kimi-k2.6"
149+ case "cerebras":
150+ return "qwen-3-235b-a22b-instruct-2507"
151+ case "groq":
152+ return "llama-3.3-70b-versatile"
153+ case "xai":
154+ return "grok-code-fast-1"
155+ case "together":
156+ return "Qwen/Qwen3-Coder-480B-A35B-Instruct"
157+ case "huggingface":
158+ return "moonshotai/Kimi-K2-Instruct"
159+ case "openrouter":
160+ return "anthropic/claude-sonnet-4.5"
161+ case "mistral":
162+ return "mistral-large-latest"
163+ case "zai":
164+ return "glm-4.7"
165+ case "xiaomi", "xiaomi-token-plan-ams", "xiaomi-token-plan-cn", "xiaomi-token-plan-sgp":
166+ return "mimo-v2.5"
167+ case "minimax", "minimax-cn":
168+ return "MiniMax-M2.7"
169+ case "fireworks":
170+ return "accounts/fireworks/models/kimi-k2p6"
171+ case "vercel-ai-gateway":
172+ return "anthropic/claude-sonnet-4.5"
173+ case "opencode":
174+ return "claude-sonnet-4-5"
175+ case "opencode-go":
176+ return "kimi-k2.6"
177+ case "amazon-bedrock":
178+ return "anthropic.claude-sonnet-4-5-20250929-v1:0"
179+ case "google-vertex":
180+ return "gemini-2.5-pro"
181+ case "azure-openai-responses":
182+ return "gpt-5"
183+ case "github-copilot":
184+ return "claude-sonnet-4.5"
145185 default:
146186 return provider.DefaultModel.ID
147187 }
148188 }
149189
190+// knownProviders is the set of provider ids zot recognises. Used by
191+// Resolve to validate args.Provider, by extension-callers, and by the
192+// auto-fallback logic that picks any logged-in provider when the user's
193+// preferred one has no credentials.
194+var knownProviders = []string{
195+ "anthropic", "openai", "openai-codex", "openai-responses", "kimi", "deepseek", "google", "ollama",
196+ "moonshotai", "moonshotai-cn",
197+ "cerebras", "groq", "xai", "together", "huggingface", "openrouter",
198+ "mistral", "zai",
199+ "xiaomi", "xiaomi-token-plan-ams", "xiaomi-token-plan-cn", "xiaomi-token-plan-sgp",
200+ "minimax", "minimax-cn",
201+ "fireworks", "vercel-ai-gateway",
202+ "opencode", "opencode-go",
203+ "amazon-bedrock", "google-vertex", "azure-openai-responses",
204+ "github-copilot", "cloudflare-workers-ai", "cloudflare-ai-gateway",
205+}
206+
207+func isKnownProvider(name string) bool {
208+ for _, p := range knownProviders {
209+ if p == name {
210+ return true
211+ }
212+ }
213+ return false
214+}
215+
150216 // Resolve merges args, config, and env into a Resolved set.
151217 //
152218 // Unlike the earlier version, Resolve NEVER returns an error for
@@ -158,7 +224,7 @@ func Resolve(args Args, requireCred bool) (Resolved, error) {
158224
159225 // User-requested provider (explicit > config > default).
160226 provName := firstNonEmpty(args.Provider, cfg.Provider, "anthropic")
161- if provName != "anthropic" && provName != "openai" && provName != "openai-codex" && provName != "kimi" && provName != "deepseek" && provName != "google" && provName != "ollama" {
227+ if !isKnownProvider(provName) {
162228 // Unknown provider (maybe removed or renamed). Fall back to
163229 // the first provider that has credentials, or anthropic.
164230 provName = "anthropic"
@@ -219,41 +285,16 @@ func Resolve(args Args, requireCred bool) (Resolved, error) {
219285
220286 model := firstNonEmpty(args.Model, cfg.Model)
221287 if model == "" {
222- switch provName {
223- case "openai":
224- model = "gpt-5"
225- case "openai-codex":
226- model = "gpt-5.5"
227- case "kimi":
228- model = "kimi-for-coding"
229- case "deepseek":
230- model = "deepseek-v4-pro"
231- case "google":
232- model = "gemini-2.5-pro"
233- case "ollama":
288+ if provName == "ollama" {
234289 return Resolved{}, fmt.Errorf("ollama requires --model (e.g. --model llama3)")
235- default:
236- model = provider.DefaultModel.ID
237290 }
291+ model = defaultModelForProvider(provName)
238292 }
239293 // If the resolved model belongs to a different provider (e.g. config
240294 // says gpt-5 but we fell back to anthropic), pick that provider's default.
241295 if provName != "ollama" {
242296 if m, err := provider.FindModel("", model); err == nil && m.Provider != provName {
243- switch provName {
244- case "openai":
245- model = "gpt-5"
246- case "openai-codex":
247- model = "gpt-5.5"
248- case "kimi":
249- model = "kimi-for-coding"
250- case "deepseek":
251- model = "deepseek-v4-pro"
252- case "google":
253- model = "gemini-2.5-pro"
254- default:
255- model = provider.DefaultModel.ID
256- }
297+ model = defaultModelForProvider(provName)
257298 }
258299 }
259300 resolvedModel, err := provider.FindModel(provName, model)
@@ -522,11 +563,17 @@ func (r Resolved) NewClient() provider.Client {
522563 case "ollama":
523564 return provider.NewOpenAI(r.Credential, r.BaseURL)
524565 case "kimi":
525- inner := provider.NewKimiWithHeaders(r.Credential, r.BaseURL, kimiCodeHeaders())
566+ // kimi-coding speaks anthropic-messages on api.kimi.com/coding.
567+ // Subscription OAuth (refreshed) wraps the same Anthropic-shaped client.
568+ inner := provider.NewKimiCodingWithHeaders(r.Credential, r.BaseURL, kimiCodeHeaders())
526569 if r.AuthMethod == "oauth" {
527570 return r.wrapWithRefresh(inner)
528571 }
529572 return inner
573+ case "moonshotai":
574+ return provider.NewMoonshot(r.Credential, r.BaseURL)
575+ case "moonshotai-cn":
576+ return provider.NewMoonshotCN(r.Credential, r.BaseURL)
530577 case "deepseek":
531578 return provider.NewDeepSeek(r.Credential, r.BaseURL)
532579 case "openai":
@@ -534,9 +581,61 @@ func (r Resolved) NewClient() provider.Client {
534581 case "openai-codex":
535582 inner := provider.NewOpenAICodex(r.Credential, r.AccountID, r.BaseURL)
536583 return r.wrapWithRefresh(inner)
584+ case "openai-responses":
585+ // Public OpenAI Responses API (api.openai.com/v1/responses) via
586+ // API key. Separate provider from `openai` (Chat Completions) and
587+ // from `openai-codex` (ChatGPT subscription OAuth).
588+ return provider.NewOpenAIResponses(r.Credential, r.BaseURL)
537589 case "google":
538- // API-key only path. Gemini Generative Language API.
539590 return provider.NewGemini(r.Credential, r.BaseURL)
591+ case "cerebras":
592+ return provider.NewCerebras(r.Credential, r.BaseURL)
593+ case "groq":
594+ return provider.NewGroq(r.Credential, r.BaseURL)
595+ case "xai":
596+ return provider.NewXAI(r.Credential, r.BaseURL)
597+ case "together":
598+ return provider.NewTogether(r.Credential, r.BaseURL)
599+ case "huggingface":
600+ return provider.NewHuggingFace(r.Credential, r.BaseURL)
601+ case "openrouter":
602+ return provider.NewOpenRouter(r.Credential, r.BaseURL)
603+ case "zai":
604+ return provider.NewZAI(r.Credential, r.BaseURL)
605+ case "xiaomi":
606+ return provider.NewXiaomi(r.Credential, r.BaseURL)
607+ case "xiaomi-token-plan-ams":
608+ return provider.NewXiaomiTokenPlan("ams", r.Credential, r.BaseURL)
609+ case "xiaomi-token-plan-cn":
610+ return provider.NewXiaomiTokenPlan("cn", r.Credential, r.BaseURL)
611+ case "xiaomi-token-plan-sgp":
612+ return provider.NewXiaomiTokenPlan("sgp", r.Credential, r.BaseURL)
613+ case "opencode":
614+ return provider.NewOpenCode(r.Credential, r.BaseURL)
615+ case "opencode-go":
616+ return provider.NewOpenCodeGo(r.Credential, r.BaseURL)
617+ case "minimax":
618+ return provider.NewMinimaxAnthropic(r.Credential, r.BaseURL)
619+ case "minimax-cn":
620+ return provider.NewMinimaxCNAnthropic(r.Credential, r.BaseURL)
621+ case "fireworks":
622+ return provider.NewFireworksAnthropic(r.Credential, r.BaseURL)
623+ case "vercel-ai-gateway":
624+ return provider.NewVercelGatewayAnthropic(r.Credential, r.BaseURL)
625+ case "mistral":
626+ return provider.NewMistral(r.Credential, r.BaseURL)
627+ case "amazon-bedrock":
628+ return provider.NewBedrock(r.Credential, r.BaseURL)
629+ case "google-vertex":
630+ return provider.NewGoogleVertex(r.Credential, r.BaseURL)
631+ case "azure-openai-responses":
632+ return provider.NewAzureOpenAIResponses(r.Credential, r.BaseURL)
633+ case "github-copilot":
634+ return provider.NewGithubCopilot(r.Credential, r.BaseURL)
635+ case "cloudflare-workers-ai":
636+ return provider.NewCloudflareWorkersAI(r.Credential, r.BaseURL)
637+ case "cloudflare-ai-gateway":
638+ return provider.NewCloudflareAIGateway(r.Credential, r.BaseURL)
540639 default:
541640 if r.AuthMethod == "oauth" {
542641 inner := provider.NewAnthropicOAuth(r.Credential, r.BaseURL)
@@ -571,7 +670,8 @@ func (r Resolved) wrapWithRefresh(inner provider.Client) provider.Client {
571670 case "openai-codex":
572671 return provider.NewOpenAICodex(token, accountID, baseURL)
573672 case "kimi":
574- return provider.NewKimiWithHeaders(token, baseURL, kimiCodeHeaders())
673+ // anthropic-messages on api.kimi.com/coding.
674+ return provider.NewKimiCodingWithHeaders(token, baseURL, kimiCodeHeaders())
575675 default:
576676 return provider.NewAnthropicOAuth(token, baseURL)
577677 }
@@ -683,7 +783,7 @@ func kimiCodeHeaders() map[string]string {
683783
684784 func envVarName(provider string) string {
685785 switch provider {
686- case "openai", "openai-codex":
786+ case "openai", "openai-codex", "openai-responses":
687787 return "OPENAI"
688788 case "kimi":
689789 return "KIMI"
@@ -693,6 +793,52 @@ func envVarName(provider string) string {
693793 return "GEMINI"
694794 case "ollama":
695795 return "OLLAMA"
796+ case "moonshotai", "moonshotai-cn":
797+ return "MOONSHOT"
798+ case "groq":
799+ return "GROQ"
800+ case "xai":
801+ return "XAI"
802+ case "cerebras":
803+ return "CEREBRAS"
804+ case "together":
805+ return "TOGETHER"
806+ case "huggingface":
807+ return "HF"
808+ case "openrouter":
809+ return "OPENROUTER"
810+ case "mistral":
811+ return "MISTRAL"
812+ case "zai":
813+ return "ZAI"
814+ case "xiaomi":
815+ return "XIAOMI"
816+ case "xiaomi-token-plan-ams":
817+ return "XIAOMI_TOKEN_PLAN_AMS"
818+ case "xiaomi-token-plan-cn":
819+ return "XIAOMI_TOKEN_PLAN_CN"
820+ case "xiaomi-token-plan-sgp":
821+ return "XIAOMI_TOKEN_PLAN_SGP"
822+ case "minimax":
823+ return "MINIMAX"
824+ case "minimax-cn":
825+ return "MINIMAX_CN"
826+ case "fireworks":
827+ return "FIREWORKS"
828+ case "vercel-ai-gateway":
829+ return "AI_GATEWAY"
830+ case "opencode", "opencode-go":
831+ return "OPENCODE"
832+ case "github-copilot":
833+ return "COPILOT_GITHUB_TOKEN"
834+ case "cloudflare-workers-ai", "cloudflare-ai-gateway":
835+ return "CLOUDFLARE"
836+ case "amazon-bedrock":
837+ return "AWS"
838+ case "google-vertex":
839+ return "GOOGLE_CLOUD"
840+ case "azure-openai-responses":
841+ return "AZURE_OPENAI"
696842 default:
697843 return "ANTHROPIC"
698844 }
M internal/agent/cli.go +55 -8
@@ -208,6 +208,13 @@ func Run(rawArgs []string, version string) error {
208208 LoadCachedModels()
209209 LoadUserModels()
210210
211+ // Repair config.json so a stale (provider, model) pair from an
212+ // interrupted /model switch can't strand the user with an
213+ // "unknown model" error on the first turn. Runs before any UI
214+ // renders so the status bar shows the post-repair pair, not the
215+ // broken one. Silent on success.
216+ ValidateAndRepairConfig()
217+
211218 if args.ListModels {
212219 printModels()
213220 return nil
@@ -660,8 +667,9 @@ func runInteractive(ctx context.Context, args Args, version string) error {
660667 }
661668 sess = newSess
662669 currentAg.SetMessages(msgs)
663- if usage, uerr := core.SessionUsage(path); uerr == nil {
664- currentAg.SeedCost(usage)
670+ if cum, last, uerr := core.SessionUsageDetail(path); uerr == nil {
671+ currentAg.SeedCost(cum)
672+ currentAg.SeedLastTurnUsage(last)
665673 }
666674 // The live agent only receives a compact resume window, but
667675 // the session file remains intact. Keep the persistence
@@ -1094,8 +1102,9 @@ func openOrCreateSession(args Args, r Resolved, ag *core.Agent, version string)
10941102 }
10951103 if s != nil {
10961104 ag.SetMessages(msgs)
1097- if usage, uerr := core.SessionUsage(s.Path); uerr == nil {
1098- ag.SeedCost(usage)
1105+ if cum, last, uerr := core.SessionUsageDetail(s.Path); uerr == nil {
1106+ ag.SeedCost(cum)
1107+ ag.SeedLastTurnUsage(last)
10991108 }
11001109 return s, nil
11011110 }
@@ -1159,8 +1168,41 @@ func readAllStdin() (string, error) {
11591168 }
11601169
11611170 func printModels() {
1162- fmt.Println("provider model id context max-out reasoning source name")
1163- for _, m := range provider.Active() {
1171+ models := provider.Active()
1172+
1173+ // Compute column widths from actual data so wide providers (e.g.
1174+ // xiaomi-token-plan-sgp) and long bedrock model ids don't force the
1175+ // `name` column off-screen. Floors mirror the historical layout so
1176+ // short catalogs look the same as before.
1177+ provW, idW, srcW := len("provider"), len("model id"), len("source")
1178+ for _, m := range models {
1179+ if w := len(m.Provider); w > provW {
1180+ provW = w
1181+ }
1182+ if w := len(m.ID); w > idW {
1183+ idW = w
1184+ }
1185+ source := m.Source
1186+ if source == "" {
1187+ source = "catalog"
1188+ }
1189+ if m.Speculative {
1190+ source = "speculative"
1191+ }
1192+ if w := len(source); w > srcW {
1193+ srcW = w
1194+ }
1195+ }
1196+
1197+ header := fmt.Sprintf("%-*s %-*s %8s %8s %s %-*s %s",
1198+ provW, "provider",
1199+ idW, "model id",
1200+ "context", "max-out", "reasoning",
1201+ srcW, "source",
1202+ "name")
1203+ fmt.Println(header)
1204+
1205+ for _, m := range models {
11641206 reason := " "
11651207 if m.Reasoning {
11661208 reason = "✓"
@@ -1172,7 +1214,12 @@ func printModels() {
11721214 if m.Speculative {
11731215 source = "speculative"
11741216 }
1175- fmt.Printf("%-10s %-30s %8d %8d %s %-11s %s\n",
1176- m.Provider, m.ID, m.ContextWindow, m.MaxOutput, reason, source, m.DisplayName)
1217+ fmt.Printf("%-*s %-*s %8d %8d %s %-*s %s\n",
1218+ provW, m.Provider,
1219+ idW, m.ID,
1220+ m.ContextWindow, m.MaxOutput,
1221+ reason,
1222+ srcW, source,
1223+ m.DisplayName)
11771224 }
11781225 }
M internal/agent/config.go +111 -0
@@ -134,6 +134,12 @@ func ResolveCredentialFull(provider, explicit string) (cred, method, accountID s
134134 }
135135 switch provider {
136136 case "anthropic":
137+ // ANTHROPIC_OAUTH_TOKEN takes precedence over ANTHROPIC_API_KEY.
138+ // Useful when both are set and the user wants subscription auth
139+ // without editing auth.json.
140+ if v := os.Getenv("ANTHROPIC_OAUTH_TOKEN"); v != "" {
141+ return v, "oauth", "", nil
142+ }
137143 if v := os.Getenv("ANTHROPIC_API_KEY"); v != "" {
138144 return v, "apikey", "", nil
139145 }
@@ -145,6 +151,12 @@ func ResolveCredentialFull(provider, explicit string) (cred, method, accountID s
145151 // ChatGPT/Codex subscription route. It intentionally ignores
146152 // OPENAI_API_KEY so users can keep both OpenAI API and Codex
147153 // subscription credentials configured and choose by provider.
154+ case "openai-responses":
155+ // Public OpenAI Responses API. Same env var as the chat-completions
156+ // `openai` provider; users pick the wire format by provider id.
157+ if v := os.Getenv("OPENAI_API_KEY"); v != "" {
158+ return v, "apikey", "", nil
159+ }
148160 case "kimi":
149161 if v := os.Getenv("KIMI_API_KEY"); v != "" {
150162 return v, "apikey", "", nil
@@ -166,6 +178,105 @@ func ResolveCredentialFull(provider, explicit string) (cred, method, accountID s
166178 if v := os.Getenv("DEEPSEEK_API_KEY"); v != "" {
167179 return v, "apikey", "", nil
168180 }
181+ case "moonshotai", "moonshotai-cn":
182+ // Moonshot direct API (separate from kimi-coding, which is the
183+ // Anthropic-Messages-fronted /coding endpoint with subscription OAuth).
184+ if v := os.Getenv("MOONSHOT_API_KEY"); v != "" {
185+ return v, "apikey", "", nil
186+ }
187+ case "groq":
188+ if v := os.Getenv("GROQ_API_KEY"); v != "" {
189+ return v, "apikey", "", nil
190+ }
191+ case "xai":
192+ if v := os.Getenv("XAI_API_KEY"); v != "" {
193+ return v, "apikey", "", nil
194+ }
195+ case "cerebras":
196+ if v := os.Getenv("CEREBRAS_API_KEY"); v != "" {
197+ return v, "apikey", "", nil
198+ }
199+ case "together":
200+ if v := os.Getenv("TOGETHER_API_KEY"); v != "" {
201+ return v, "apikey", "", nil
202+ }
203+ case "huggingface":
204+ if v := os.Getenv("HF_TOKEN"); v != "" {
205+ return v, "apikey", "", nil
206+ }
207+ case "openrouter":
208+ if v := os.Getenv("OPENROUTER_API_KEY"); v != "" {
209+ return v, "apikey", "", nil
210+ }
211+ case "mistral":
212+ if v := os.Getenv("MISTRAL_API_KEY"); v != "" {
213+ return v, "apikey", "", nil
214+ }
215+ case "zai":
216+ if v := os.Getenv("ZAI_API_KEY"); v != "" {
217+ return v, "apikey", "", nil
218+ }
219+ case "xiaomi", "xiaomi-token-plan-ams", "xiaomi-token-plan-cn", "xiaomi-token-plan-sgp":
220+ envVar := "XIAOMI_API_KEY"
221+ switch provider {
222+ case "xiaomi-token-plan-ams":
223+ envVar = "XIAOMI_TOKEN_PLAN_AMS_API_KEY"
224+ case "xiaomi-token-plan-cn":
225+ envVar = "XIAOMI_TOKEN_PLAN_CN_API_KEY"
226+ case "xiaomi-token-plan-sgp":
227+ envVar = "XIAOMI_TOKEN_PLAN_SGP_API_KEY"
228+ }
229+ if v := os.Getenv(envVar); v != "" {
230+ return v, "apikey", "", nil
231+ }
232+ case "minimax":
233+ if v := os.Getenv("MINIMAX_API_KEY"); v != "" {
234+ return v, "apikey", "", nil
235+ }
236+ case "minimax-cn":
237+ if v := os.Getenv("MINIMAX_CN_API_KEY"); v != "" {
238+ return v, "apikey", "", nil
239+ }
240+ if v := os.Getenv("MINIMAX_API_KEY"); v != "" {
241+ return v, "apikey", "", nil
242+ }
243+ case "fireworks":
244+ if v := os.Getenv("FIREWORKS_API_KEY"); v != "" {
245+ return v, "apikey", "", nil
246+ }
247+ case "vercel-ai-gateway":
248+ if v := os.Getenv("AI_GATEWAY_API_KEY"); v != "" {
249+ return v, "apikey", "", nil
250+ }
251+ case "opencode", "opencode-go":
252+ if v := os.Getenv("OPENCODE_API_KEY"); v != "" {
253+ return v, "apikey", "", nil
254+ }
255+ case "github-copilot":
256+ if v := os.Getenv("COPILOT_GITHUB_TOKEN"); v != "" {
257+ return v, "apikey", "", nil
258+ }
259+ case "cloudflare-workers-ai", "cloudflare-ai-gateway":
260+ if v := os.Getenv("CLOUDFLARE_API_KEY"); v != "" {
261+ return v, "apikey", "", nil
262+ }
263+ case "amazon-bedrock":
264+ // Bedrock has many credential sources (AWS_PROFILE, IAM keys,
265+ // container creds, IRSA, bearer token). We surface a sentinel so
266+ // Resolve doesn't error on missing key; the real client (when
267+ // implemented) will resolve credentials through aws-sdk-go-v2.
268+ if os.Getenv("AWS_ACCESS_KEY_ID") != "" || os.Getenv("AWS_PROFILE") != "" ||
269+ os.Getenv("AWS_BEARER_TOKEN_BEDROCK") != "" {
270+ return "<aws>", "apikey", "", nil
271+ }
272+ case "google-vertex":
273+ if v := os.Getenv("GOOGLE_CLOUD_API_KEY"); v != "" {
274+ return v, "apikey", "", nil
275+ }
276+ case "azure-openai-responses":
277+ if v := os.Getenv("AZURE_OPENAI_API_KEY"); v != "" {
278+ return v, "apikey", "", nil
279+ }
169280 }
170281 c, err := AuthStoreFor().Load()
171282 if err != nil {
M internal/agent/modelsync.go +65 -1
@@ -2,6 +2,8 @@ package agent
22
33 import (
44 "context"
5+ "fmt"
6+ "os"
57 "path/filepath"
68 "time"
79
@@ -33,11 +35,73 @@ func LoadCachedModels() {
3335
3436 // LoadUserModels reads $ZOT_HOME/models.json and merges any user-defined
3537 // models into the active catalog. User models take highest precedence.
38+// Any validation issues (bad provider id, empty model id, malformed
39+// JSON, negative widths) are surfaced as one warning per line on stderr;
40+// the well-formed entries from the rest of the file are still loaded.
3641 func LoadUserModels() {
37- models := provider.LoadUserModels(UserModelsPath())
42+ models, warnings := provider.LoadUserModelsWithWarnings(UserModelsPath())
43+ for _, w := range warnings {
44+ fmt.Fprintln(os.Stderr, "zot:", w)
45+ }
3846 provider.SetUserModels(models)
3947 }
4048
49+// ValidateAndRepairConfig checks the persisted config.json's
50+// (Provider, Model) pair against the active catalog and repairs any
51+// mismatch in-place (and on disk) before any UI renders. Three failure
52+// modes are handled:
53+//
54+// - cfg.Provider is empty or unknown -> reset to "anthropic".
55+// - cfg.Model is empty -> set to the provider's default.
56+// - cfg.Model belongs to a different provider than cfg.Provider
57+// (e.g. provider=anthropic + model=kimi-for-coding from a stale
58+// half-applied switch) -> reset model to the provider's default.
59+//
60+// Silent on success; one stderr line per repair. Errors loading or
61+// saving the file are non-fatal — the caller continues with defaults.
62+func ValidateAndRepairConfig() {
63+ cfg, err := LoadConfig()
64+ if err != nil {
65+ fmt.Fprintf(os.Stderr, "zot: config.json: %v (using defaults)\n", err)
66+ return
67+ }
68+ changed := false
69+
70+ if cfg.Provider != "" && !isKnownProvider(cfg.Provider) {
71+ fmt.Fprintf(os.Stderr, "zot: config.json: unknown provider %q reset to \"anthropic\"\n", cfg.Provider)
72+ cfg.Provider = "anthropic"
73+ cfg.Model = ""
74+ changed = true
75+ }
76+
77+ if cfg.Provider != "" && cfg.Model != "" {
78+ if m, err := provider.FindModel("", cfg.Model); err == nil {
79+ if m.Provider != cfg.Provider {
80+ fix := defaultModelForProvider(cfg.Provider)
81+ fmt.Fprintf(os.Stderr,
82+ "zot: config.json: model %q belongs to provider %q (config has provider=%q); switched model to %q\n",
83+ cfg.Model, m.Provider, cfg.Provider, fix)
84+ cfg.Model = fix
85+ changed = true
86+ }
87+ } else if cfg.Provider != "ollama" {
88+ // Model id not in any catalog. Reset to provider's default.
89+ fix := defaultModelForProvider(cfg.Provider)
90+ fmt.Fprintf(os.Stderr,
91+ "zot: config.json: model %q not found in the active catalog; switched to %q\n",
92+ cfg.Model, fix)
93+ cfg.Model = fix
94+ changed = true
95+ }
96+ }
97+
98+ if changed {
99+ if err := SaveConfig(cfg); err != nil {
100+ fmt.Fprintf(os.Stderr, "zot: config.json: failed to persist repair: %v\n", err)
101+ }
102+ }
103+}
104+
41105 // RefreshModelsAsync kicks a background discovery for every provider
42106 // we have credentials for. Refreshed results are merged into the
43107 // active catalog and persisted to the on-disk cache.
A internal/agent/modelsync_test.go +103 -0
@@ -0,0 +1,103 @@
1+package agent
2+
3+import (
4+ "encoding/json"
5+ "os"
6+ "path/filepath"
7+ "testing"
8+)
9+
10+// TestValidateAndRepairConfig_MismatchedPair simulates the bug from a
11+// stale /model switch: provider=anthropic but model=kimi-for-coding
12+// (which belongs to provider=kimi). The validator should rewrite the
13+// model to anthropic's default and persist.
14+func TestValidateAndRepairConfig_MismatchedPair(t *testing.T) {
15+ home := t.TempDir()
16+ t.Setenv("ZOT_HOME", home)
17+
18+ must := func(c Config) {
19+ t.Helper()
20+ b, _ := json.Marshal(c)
21+ if err := os.WriteFile(filepath.Join(home, "config.json"), b, 0o644); err != nil {
22+ t.Fatal(err)
23+ }
24+ }
25+ must(Config{Provider: "anthropic", Model: "kimi-for-coding"})
26+
27+ ValidateAndRepairConfig()
28+
29+ out, err := LoadConfig()
30+ if err != nil {
31+ t.Fatal(err)
32+ }
33+ if out.Provider != "anthropic" {
34+ t.Errorf("provider not preserved: %q", out.Provider)
35+ }
36+ if out.Model == "kimi-for-coding" {
37+ t.Errorf("model not repaired; still %q", out.Model)
38+ }
39+ if out.Model == "" {
40+ t.Errorf("model not set; expected anthropic default")
41+ }
42+}
43+
44+// TestValidateAndRepairConfig_UnknownProvider resets to anthropic and
45+// clears the model when the saved provider id isn't recognised
46+// (e.g. user removed it from a previous build).
47+func TestValidateAndRepairConfig_UnknownProvider(t *testing.T) {
48+ home := t.TempDir()
49+ t.Setenv("ZOT_HOME", home)
50+
51+ b, _ := json.Marshal(Config{Provider: "made-up-provider", Model: "some-model"})
52+ _ = os.WriteFile(filepath.Join(home, "config.json"), b, 0o644)
53+
54+ ValidateAndRepairConfig()
55+
56+ out, _ := LoadConfig()
57+ if out.Provider != "anthropic" {
58+ t.Errorf("provider not reset: %q", out.Provider)
59+ }
60+ if out.Model != "" {
61+ t.Errorf("model not cleared: %q", out.Model)
62+ }
63+}
64+
65+// TestValidateAndRepairConfig_UnknownModel keeps the provider but
66+// snaps the model to that provider's default when the saved id is no
67+// longer in the catalog.
68+func TestValidateAndRepairConfig_UnknownModel(t *testing.T) {
69+ home := t.TempDir()
70+ t.Setenv("ZOT_HOME", home)
71+
72+ b, _ := json.Marshal(Config{Provider: "anthropic", Model: "claude-deleted-model"})
73+ _ = os.WriteFile(filepath.Join(home, "config.json"), b, 0o644)
74+
75+ ValidateAndRepairConfig()
76+
77+ out, _ := LoadConfig()
78+ if out.Provider != "anthropic" {
79+ t.Errorf("provider changed: %q", out.Provider)
80+ }
81+ if out.Model == "" || out.Model == "claude-deleted-model" {
82+ t.Errorf("model not repaired: %q", out.Model)
83+ }
84+}
85+
86+// TestValidateAndRepairConfig_HappyPath leaves a valid config alone.
87+func TestValidateAndRepairConfig_HappyPath(t *testing.T) {
88+ home := t.TempDir()
89+ t.Setenv("ZOT_HOME", home)
90+
91+ b, _ := json.Marshal(Config{Provider: "anthropic", Model: "claude-sonnet-4-5"})
92+ _ = os.WriteFile(filepath.Join(home, "config.json"), b, 0o644)
93+
94+ ValidateAndRepairConfig()
95+
96+ out, _ := LoadConfig()
97+ if out.Provider != "anthropic" {
98+ t.Errorf("provider mutated: %q", out.Provider)
99+ }
100+ if out.Model != "claude-sonnet-4-5" {
101+ t.Errorf("model mutated: %q", out.Model)
102+ }
103+}
M internal/agent/modes/interactive.go +66 -0
@@ -203,6 +203,7 @@ type chatCacheKey struct {
203203 updateURL string
204204 welcomeShowVer bool
205205 expandAll bool
206+ tailLimit int
206207 }
207208
208209 // SettingsStore persists user-toggleable settings surfaced by /settings.
@@ -367,6 +368,23 @@ type Interactive struct {
367368 // enough to read at a glance and keeps the splash short.
368369 const welcomeVersionDuration = 1500 * time.Millisecond
369370
371+// initialResumeTailLimit caps how many messages from a freshly-resumed
372+// transcript we render on the first paint. The full transcript is
373+// still in memory; older messages are rendered (and their cached
374+// lines kept for the lifetime of the View) as soon as the user
375+// scrolls past the rendered tail. Picked to comfortably fill the
376+// largest realistic terminal viewport while keeping first paint
377+// snappy on multi-thousand-message sessions where markdown / syntax
378+// highlighting dominates the redraw cost.
379+const initialResumeTailLimit = 80
380+
381+// resumeTailExpandStep is how many additional messages the tail
382+// limit grows by each time the user scrolls past the currently
383+// rendered top. Pre-rendering this many messages on a single tick
384+// keeps scroll-up smooth without falling back to a one-by-one
385+// reveal that would feel jerky.
386+const resumeTailExpandStep = 80
387+
370388 // NewInteractive constructs an Interactive from cfg.
371389 func NewInteractive(cfg InteractiveConfig) *Interactive {
372390 i := &Interactive{
@@ -407,6 +425,18 @@ func NewInteractive(cfg InteractiveConfig) *Interactive {
407425 i.agent = cfg.Agent
408426 i.view.Messages = cfg.Agent.Messages()
409427 i.cumUsage = cfg.Agent.Cost()
428+ // Rehydrate the "context used" gauge from the last persisted
429+ // turn. Without this the status bar reads 0.0% after a resume
430+ // until the next turn lands a usage event.
431+ if last := cfg.Agent.LastTurnUsage(); last.InputTokens > 0 || last.CacheReadTokens > 0 || last.CacheWriteTokens > 0 {
432+ i.lastCtxInput = last.InputTokens + last.CacheReadTokens + last.CacheWriteTokens
433+ }
434+ // Cap the first paint at the tail of the transcript so
435+ // resuming a multi-thousand-message session doesn't block
436+ // on rendering every prior turn before showing anything.
437+ if len(i.view.Messages) > initialResumeTailLimit {
438+ i.view.TailLimit = initialResumeTailLimit
439+ }
410440 }
411441 return i
412442 }
@@ -700,6 +730,7 @@ func (i *Interactive) chatCacheKeyLocked(cols int) (chatCacheKey, bool) {
700730 updateURL: i.updateInfo.URL,
701731 welcomeShowVer: showVer,
702732 expandAll: i.view.ExpandAll,
733+ tailLimit: i.view.TailLimit,
703734 }, true
704735 }
705736
@@ -1114,6 +1145,28 @@ func (i *Interactive) redraw() {
11141145 if maxOffset < 0 {
11151146 maxOffset = 0
11161147 }
1148+ // Tail-render expansion: if the user has scrolled to (or above)
1149+ // the top of the currently rendered tail and there are still
1150+ // truncated messages above, widen view.TailLimit and rebuild.
1151+ // The chat cache is keyed on tailLimit so the next cachedChatLocked
1152+ // will re-issue Build instead of returning the stale slice. We
1153+ // rebuild immediately so the same redraw shows the freshly-revealed
1154+ // rows; otherwise the user would have to scroll again to see them.
1155+ if i.scrollOffset >= maxOffset && i.view.TailLimit > 0 && i.view.TailLimit < len(i.view.Messages) {
1156+ i.view.TailLimit += resumeTailExpandStep
1157+ if i.view.TailLimit >= len(i.view.Messages) {
1158+ i.view.TailLimit = 0 // unbounded
1159+ }
1160+ i.chatCacheValid = false
1161+ chat = i.cachedChatLocked(cols)
1162+ for len(chat) > 0 && strings.TrimSpace(chat[len(chat)-1]) == "" {
1163+ chat = chat[:len(chat)-1]
1164+ }
1165+ maxOffset = len(chat) - chatRows
1166+ if maxOffset < 0 {
1167+ maxOffset = 0
1168+ }
1169+ }
11171170 if i.scrollOffset > maxOffset {
11181171 i.scrollOffset = maxOffset
11191172 }
@@ -3279,6 +3332,19 @@ func (i *Interactive) applySessionSelection(path string) {
32793332 if i.agent != nil {
32803333 i.view.Messages = i.agent.Messages()
32813334 i.cumUsage = i.agent.Cost()
3335+ if last := i.agent.LastTurnUsage(); last.InputTokens > 0 || last.CacheReadTokens > 0 || last.CacheWriteTokens > 0 {
3336+ i.lastCtxInput = last.InputTokens + last.CacheReadTokens + last.CacheWriteTokens
3337+ } else {
3338+ i.lastCtxInput = 0
3339+ }
3340+ // Snap to the tail again — the swap brought in a fresh
3341+ // transcript whose markdown / chroma cost we don't want
3342+ // blocking the redraw.
3343+ if len(i.view.Messages) > initialResumeTailLimit {
3344+ i.view.TailLimit = initialResumeTailLimit
3345+ } else {
3346+ i.view.TailLimit = 0
3347+ }
32823348 }
32833349 i.invalidate()
32843350 }()
M internal/agent/modes/model_dialog.go +50 -4
@@ -5,6 +5,7 @@ import (
55 "sort"
66 "strings"
77
8+ "github.com/mattn/go-runewidth"
89 "github.com/patriceckhart/zot/internal/provider"
910 "github.com/patriceckhart/zot/internal/tui"
1011 )
@@ -22,6 +23,13 @@ type modelDialog struct {
2223 cursor int
2324 current string // currently selected model id (highlighted)
2425 query string // live filter text typed by the user
26+
27+ // Column widths are computed once on Open() across the entire
28+ // catalog so the layout stays stable while the user scrolls or
29+ // filters. Recomputing per visible window would make the columns
30+ // jitter left/right whenever rows entered or left the 14-row band.
31+ provW int
32+ idW int
2533 }
2634
2735 // modelDialogAction is returned by HandleKey.
@@ -57,6 +65,7 @@ func (d *modelDialog) Open(current string, loggedInProviders []string) {
5765 d.all = sortedModels(all)
5866 d.current = current
5967 d.query = ""
68+ d.provW, d.idW = columnWidths(d.all)
6069 d.refilter()
6170 }
6271
@@ -159,15 +168,17 @@ func (d *modelDialog) Render(th tui.Theme, width int) []string {
159168 end = start + visible
160169 }
161170
171+ // Column widths were computed once on Open() across the full catalog
172+ // (d.provW / d.idW). Reusing them keeps the layout rock-stable while
173+ // scrolling and filtering — recomputing per visible window made the
174+ // columns visibly jitter.
175+ provW, idW := d.provW, d.idW
162176 for i := start; i < end; i++ {
163177 m := d.view[i]
164- prov := m.Provider
165- id := m.ID
166178 reason := " "
167179 if m.Reasoning {
168180 reason = "✦"
169181 }
170- name := m.DisplayName
171182 tag := ""
172183 switch {
173184 case m.Speculative:
@@ -179,7 +190,11 @@ func (d *modelDialog) Render(th tui.Theme, width int) []string {
179190 if m.ID == d.current {
180191 curMark = "● "
181192 }
182- plain := fmt.Sprintf(" %s%-10s %-28s %s %s%s", curMark, prov, id, reason, tag, name)
193+ plain := fmt.Sprintf(" %s%s %s %s %s%s",
194+ curMark,
195+ padRight(m.Provider, provW),
196+ padRight(m.ID, idW),
197+ reason, tag, m.DisplayName)
183198 if i == d.cursor {
184199 lines = append(lines, th.PadHighlight(plain, width))
185200 } else {
@@ -198,6 +213,37 @@ func (d *modelDialog) Render(th tui.Theme, width int) []string {
198213 return lines
199214 }
200215
216+// columnWidths returns the display-cell width of the longest provider
217+// name and the longest model id in rows. Each column is clamped to a
218+// minimum so a single-row filter still renders sensibly.
219+func columnWidths(rows []provider.Model) (provW, idW int) {
220+ const minProv, minID = 6, 12
221+ provW, idW = minProv, minID
222+ for _, m := range rows {
223+ if w := runewidth.StringWidth(m.Provider); w > provW {
224+ provW = w
225+ }
226+ if w := runewidth.StringWidth(m.ID); w > idW {
227+ idW = w
228+ }
229+ }
230+ return
231+}
232+
233+// padRight returns s padded with spaces on the right so its display
234+// width equals w. Truncates with an ellipsis when s is too wide to
235+// avoid blowing out the column.
236+func padRight(s string, w int) string {
237+ cw := runewidth.StringWidth(s)
238+ if cw == w {
239+ return s
240+ }
241+ if cw < w {
242+ return s + strings.Repeat(" ", w-cw)
243+ }
244+ return runewidth.Truncate(s, w, "…")
245+}
246+
201247 // HandleKey advances the dialog and returns an action to apply, if any.
202248 func (d *modelDialog) HandleKey(k tui.Key) modelDialogAction {
203249 switch k.Kind {
M internal/auth/probe.go +63 -0
@@ -52,6 +52,69 @@ func ProbeAPIKey(ctx context.Context, provider, key string) error {
5252 return err
5353 }
5454 req.Header.Set("x-goog-api-key", key)
55+ // OpenAI-compatible third parties: a GET /v1/models with bearer auth
56+ // is enough to validate the key. Branches kept explicit (rather than a
57+ // generic default) so the URL list is searchable and reviewable.
58+ case "moonshotai":
59+ req, err = http.NewRequestWithContext(ctx, "GET", "https://api.moonshot.ai/v1/models", nil)
60+ if err != nil {
61+ return err
62+ }
63+ req.Header.Set("authorization", "Bearer "+key)
64+ case "moonshotai-cn":
65+ req, err = http.NewRequestWithContext(ctx, "GET", "https://api.moonshot.cn/v1/models", nil)
66+ if err != nil {
67+ return err
68+ }
69+ req.Header.Set("authorization", "Bearer "+key)
70+ case "groq":
71+ req, err = http.NewRequestWithContext(ctx, "GET", "https://api.groq.com/openai/v1/models", nil)
72+ if err != nil {
73+ return err
74+ }
75+ req.Header.Set("authorization", "Bearer "+key)
76+ case "cerebras":
77+ req, err = http.NewRequestWithContext(ctx, "GET", "https://api.cerebras.ai/v1/models", nil)
78+ if err != nil {
79+ return err
80+ }
81+ req.Header.Set("authorization", "Bearer "+key)
82+ case "xai":
83+ req, err = http.NewRequestWithContext(ctx, "GET", "https://api.x.ai/v1/models", nil)
84+ if err != nil {
85+ return err
86+ }
87+ req.Header.Set("authorization", "Bearer "+key)
88+ case "together":
89+ req, err = http.NewRequestWithContext(ctx, "GET", "https://api.together.ai/v1/models", nil)
90+ if err != nil {
91+ return err
92+ }
93+ req.Header.Set("authorization", "Bearer "+key)
94+ case "openrouter":
95+ req, err = http.NewRequestWithContext(ctx, "GET", "https://openrouter.ai/api/v1/models", nil)
96+ if err != nil {
97+ return err
98+ }
99+ req.Header.Set("authorization", "Bearer "+key)
100+ case "huggingface":
101+ req, err = http.NewRequestWithContext(ctx, "GET", "https://router.huggingface.co/v1/models", nil)
102+ if err != nil {
103+ return err
104+ }
105+ req.Header.Set("authorization", "Bearer "+key)
106+ case "zai":
107+ req, err = http.NewRequestWithContext(ctx, "GET", "https://api.z.ai/api/coding/paas/v4/models", nil)
108+ if err != nil {
109+ return err
110+ }
111+ req.Header.Set("authorization", "Bearer "+key)
112+ case "mistral":
113+ req, err = http.NewRequestWithContext(ctx, "GET", "https://api.mistral.ai/v1/models", nil)
114+ if err != nil {
115+ return err
116+ }
117+ req.Header.Set("authorization", "Bearer "+key)
55118 default:
56119 return fmt.Errorf("unknown provider %q", provider)
57120 }
M internal/core/agent.go +18 -0
@@ -236,6 +236,24 @@ func (a *Agent) SeedCost(u provider.Usage) {
236236 a.cost.Total = u
237237 }
238238
239+// LastTurnUsage returns the per-turn usage of the most recent
240+// completed turn. Drives the "context used" gauge in the status bar
241+// without waiting for the next turn to land.
242+func (a *Agent) LastTurnUsage() provider.Usage {
243+ a.mu.Lock()
244+ defer a.mu.Unlock()
245+ return a.cost.LastTurn
246+}
247+
248+// SeedLastTurnUsage primes the per-turn snapshot. Used on resume so
249+// the gauge reflects the prompt size of the last turn in the session
250+// file instead of starting at zero.
251+func (a *Agent) SeedLastTurnUsage(u provider.Usage) {
252+ a.mu.Lock()
253+ defer a.mu.Unlock()
254+ a.cost.LastTurn = u
255+}
256+
239257 // fireMessageAppended invokes OnMessageAppended without holding the
240258 // agent mutex, so the host's persistence callback can take its own
241259 // locks without deadlocking the agent loop. Tolerates a nil hook so
M internal/core/cost.go +11 -2
@@ -3,12 +3,21 @@ package core
33 import "github.com/patriceckhart/zot/internal/provider"
44
55 // CostTracker accumulates usage across turns in a session.
6+//
7+// Total is the cumulative usage shown in the status bar's "$x.xx"
8+// readout. LastTurn is the per-turn usage of the most recent
9+// completed turn; the TUI uses LastTurn.InputTokens+cache as a
10+// proxy for "current context size" so the X%/Ymax gauge tracks the
11+// prompt size that just went to the model.
612 type CostTracker struct {
7- Total provider.Usage
13+ Total provider.Usage
14+ LastTurn provider.Usage
815 }
916
10-// Add folds u into the running total and returns the new cumulative value.
17+// Add folds u into the running total, records u as the last-turn
18+// snapshot, and returns the new cumulative value.
1119 func (c *CostTracker) Add(u provider.Usage) provider.Usage {
1220 c.Total = c.Total.Add(u)
21+ c.LastTurn = u
1322 return c.Total
1423 }
M internal/core/session.go +60 -10
@@ -160,14 +160,33 @@ func forEachJSONLLine(r io.Reader, fn func([]byte) error) error {
160160 // latest row's cumulative field is the session total. Missing usage rows
161161 // are valid for old/empty sessions and return the zero value.
162162 func SessionUsage(path string) (provider.Usage, error) {
163- f, err := os.Open(path)
164- if err != nil {
165- return provider.Usage{}, err
163+ cum, _, err := SessionUsageDetail(path)
164+ return cum, err
165+}
166+
167+// SessionUsageDetail returns the latest cumulative usage and the
168+// per-turn usage of the final completed turn. The per-turn row drives
169+// the live "context used" gauge in the status bar (input + cache
170+// approximates the prompt size the model just saw), letting the TUI
171+// rehydrate the gauge on resume instead of starting at 0% until the
172+// next turn lands.
173+func SessionUsageDetail(path string) (cumulative, lastTurn provider.Usage, err error) {
174+ f, ferr := os.Open(path)
175+ if ferr != nil {
176+ return provider.Usage{}, provider.Usage{}, ferr
166177 }
167178 defer f.Close()
168179
169- var usage provider.Usage
170- if err := forEachJSONLLine(f, func(line []byte) error {
180+ // Some historical sessions logged the per-turn `usage` field as a copy
181+ // of `cumulative` instead of the true delta. To recover an accurate
182+ // last-turn snapshot (used by the status-bar context gauge on resume),
183+ // we always derive lastTurn from the delta between the final two
184+ // cumulative rows. For prompt-size purposes, cache_read/cache_write
185+ // reflect the most recent prompt directly, so we take those from the
186+ // final cumulative row as-is rather than as a delta.
187+ var prevCum provider.Usage
188+ var haveCum bool
189+ if ierr := forEachJSONLLine(f, func(line []byte) error {
171190 var head sessionLineHead
172191 if err := json.Unmarshal(line, &head); err != nil || head.Type != "usage" {
173192 return nil
@@ -175,14 +194,45 @@ func SessionUsage(path string) (provider.Usage, error) {
175194 var row struct {
176195 Cumulative provider.Usage `json:"cumulative"`
177196 }
178- if err := json.Unmarshal(line, &row); err == nil {
179- usage = row.Cumulative
197+ if err := json.Unmarshal(line, &row); err != nil {
198+ return nil
199+ }
200+ if haveCum {
201+ prevCum = cumulative
180202 }
203+ cumulative = row.Cumulative
204+ haveCum = true
181205 return nil
182- }); err != nil {
183- return provider.Usage{}, err
206+ }); ierr != nil {
207+ return provider.Usage{}, provider.Usage{}, ierr
208+ }
209+ if haveCum {
210+ // input/output are monotonic totals -> per-turn = delta.
211+ lastTurn.InputTokens = nonNegDelta(cumulative.InputTokens, prevCum.InputTokens)
212+ lastTurn.OutputTokens = nonNegDelta(cumulative.OutputTokens, prevCum.OutputTokens)
213+ // cache_read/write on the final row already represent the last prompt's
214+ // cache hit/creation, not a running total of bytes; use directly.
215+ lastTurn.CacheReadTokens = cumulative.CacheReadTokens - prevCum.CacheReadTokens
216+ if lastTurn.CacheReadTokens < 0 {
217+ lastTurn.CacheReadTokens = cumulative.CacheReadTokens
218+ }
219+ lastTurn.CacheWriteTokens = cumulative.CacheWriteTokens - prevCum.CacheWriteTokens
220+ if lastTurn.CacheWriteTokens < 0 {
221+ lastTurn.CacheWriteTokens = cumulative.CacheWriteTokens
222+ }
223+ lastTurn.CostUSD = cumulative.CostUSD - prevCum.CostUSD
224+ if lastTurn.CostUSD < 0 {
225+ lastTurn.CostUSD = 0
226+ }
227+ }
228+ return cumulative, lastTurn, nil
229+}
230+
231+func nonNegDelta(cur, prev int) int {
232+ if cur < prev {
233+ return cur
184234 }
185- return usage, nil
235+ return cur - prev
186236 }
187237
188238 // OpenSession opens an existing session for appending.
A internal/provider/amazon_bedrock.go +738 -0
@@ -0,0 +1,738 @@
1+package provider
2+
3+// Amazon Bedrock Converse-Stream client.
4+//
5+// Endpoint: POST https://bedrock-runtime.{region}.amazonaws.com/model/{id}/converse-stream
6+//
7+// Auth: two paths supported.
8+//
9+// - **Bearer token** via AWS_BEARER_TOKEN_BEDROCK. Modern, simple,
10+// `bedrock:CallWithBearerToken` IAM permission required.
11+// - **SigV4 signing** via AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY
12+// (+ optional AWS_SESSION_TOKEN for temporary credentials, e.g. SSO
13+// or IRSA-issued). Hand-rolled here so we don't pull in aws-sdk-go-v2.
14+//
15+// If both are present, bearer wins (simpler / fewer moving parts).
16+//
17+// Wire format: AWS event-stream binary framing. Each message is:
18+//
19+// [4-byte total length BE]
20+// [4-byte headers length BE]
21+// [4-byte CRC32 of the 8-byte prelude]
22+// [headers ...] each = [1B name-length][name][1B value-type][2B value-length][value]
23+// [payload ...] payload-length = total - headers - 4 (trailer) - 12 (prelude+prelude-crc)
24+// [4-byte CRC32 of message bytes minus trailer]
25+//
26+// We only consume the stream — never produce it — so we don't validate
27+// CRCs (the TLS+TCP transport already protects in-flight bytes). We do
28+// validate length-bounds to avoid OOB reads on a hostile / corrupted
29+// stream.
30+//
31+// Event types we care about (from `:event-type` header):
32+// - messageStart { role: "assistant" }
33+// - contentBlockStart { contentBlockIndex, start: { toolUse: { toolUseId, name } } }
34+// - contentBlockDelta { contentBlockIndex, delta: { text?, toolUse?: { input } } }
35+// - contentBlockStop { contentBlockIndex }
36+// - messageStop { stopReason }
37+// - metadata { usage: { inputTokens, outputTokens } }
38+
39+import (
40+ "bytes"
41+ "context"
42+ "crypto/hmac"
43+ "crypto/sha256"
44+ "encoding/binary"
45+ "encoding/hex"
46+ "encoding/json"
47+ "fmt"
48+ "io"
49+ "net/http"
50+ "net/url"
51+ "os"
52+ "sort"
53+ "strings"
54+ "time"
55+)
56+
57+type bedrockClient struct {
58+ // Auth mode is determined at construction time. Exactly one of
59+ // bearerToken or sigv4 is populated.
60+ bearerToken string
61+ sigv4 *bedrockSigV4Creds
62+
63+ region string
64+ baseURL string
65+ http *http.Client
66+}
67+
68+// bedrockSigV4Creds holds the IAM credentials used to sign each request.
69+type bedrockSigV4Creds struct {
70+ accessKeyID string
71+ secretAccessKey string
72+ sessionToken string // optional; STS / SSO / IRSA temp creds
73+}
74+
75+// NewBedrockClient returns a Bedrock client.
76+//
77+// Auth resolution (first match wins):
78+//
79+// 1. apiKey == real bearer-ish string (not "<aws>") -> bearer route.
80+// 2. AWS_BEARER_TOKEN_BEDROCK env var -> bearer route.
81+// 3. AWS_ACCESS_KEY_ID + AWS_SECRET_ACCESS_KEY (+ optional
82+// AWS_SESSION_TOKEN) -> SigV4 route.
83+// 4. AWS_PROFILE -> read ~/.aws/credentials, take that profile's keys.
84+//
85+// region defaults to us-east-1 unless AWS_REGION / AWS_DEFAULT_REGION is
86+// set or the baseURL embeds a region.
87+func NewBedrockClient(apiKey, baseURL string) Client {
88+ region := os.Getenv("AWS_REGION")
89+ if region == "" {
90+ region = os.Getenv("AWS_DEFAULT_REGION")
91+ }
92+ if region == "" {
93+ region = "us-east-1"
94+ }
95+ if baseURL == "" {
96+ baseURL = "https://bedrock-runtime." + region + ".amazonaws.com"
97+ }
98+ c := &bedrockClient{
99+ region: region,
100+ baseURL: strings.TrimRight(baseURL, "/"),
101+ http: &http.Client{Timeout: 0},
102+ }
103+
104+ // Bearer route.
105+ token := apiKey
106+ if token == "" || token == "<aws>" {
107+ token = os.Getenv("AWS_BEARER_TOKEN_BEDROCK")
108+ }
109+ if token != "" && token != "<aws>" {
110+ c.bearerToken = token
111+ return c
112+ }
113+
114+ // SigV4 route: env vars first.
115+ ak := os.Getenv("AWS_ACCESS_KEY_ID")
116+ sk := os.Getenv("AWS_SECRET_ACCESS_KEY")
117+ st := os.Getenv("AWS_SESSION_TOKEN")
118+ if ak != "" && sk != "" {
119+ c.sigv4 = &bedrockSigV4Creds{accessKeyID: ak, secretAccessKey: sk, sessionToken: st}
120+ return c
121+ }
122+
123+ // SigV4 route: ~/.aws/credentials via AWS_PROFILE.
124+ if profile := os.Getenv("AWS_PROFILE"); profile != "" {
125+ if creds, err := readAWSCredentialsFile(profile); err == nil {
126+ c.sigv4 = creds
127+ return c
128+ }
129+ }
130+
131+ return &unimplementedClient{
132+ name: "amazon-bedrock",
133+ hint: "no Bedrock credentials found (set AWS_BEARER_TOKEN_BEDROCK, AWS_ACCESS_KEY_ID+AWS_SECRET_ACCESS_KEY, or AWS_PROFILE)",
134+ }
135+}
136+
137+// readAWSCredentialsFile parses ~/.aws/credentials and returns the
138+// access-key/secret-key (and optional session-token) for the named
139+// profile. The file is an INI-like format with `[profile]` headers.
140+func readAWSCredentialsFile(profile string) (*bedrockSigV4Creds, error) {
141+ home, err := os.UserHomeDir()
142+ if err != nil {
143+ return nil, err
144+ }
145+ b, err := os.ReadFile(home + "/.aws/credentials")
146+ if err != nil {
147+ return nil, err
148+ }
149+ var current string
150+ creds := map[string]map[string]string{}
151+ for _, line := range strings.Split(string(b), "\n") {
152+ line = strings.TrimSpace(line)
153+ if line == "" || strings.HasPrefix(line, "#") {
154+ continue
155+ }
156+ if strings.HasPrefix(line, "[") && strings.HasSuffix(line, "]") {
157+ current = strings.TrimSpace(line[1 : len(line)-1])
158+ creds[current] = map[string]string{}
159+ continue
160+ }
161+ if current == "" {
162+ continue
163+ }
164+ eq := strings.IndexByte(line, '=')
165+ if eq < 0 {
166+ continue
167+ }
168+ k := strings.TrimSpace(line[:eq])
169+ v := strings.TrimSpace(line[eq+1:])
170+ creds[current][k] = v
171+ }
172+ p, ok := creds[profile]
173+ if !ok {
174+ return nil, fmt.Errorf("aws profile %q not found in ~/.aws/credentials", profile)
175+ }
176+ ak := p["aws_access_key_id"]
177+ sk := p["aws_secret_access_key"]
178+ if ak == "" || sk == "" {
179+ return nil, fmt.Errorf("aws profile %q missing aws_access_key_id or aws_secret_access_key", profile)
180+ }
181+ return &bedrockSigV4Creds{
182+ accessKeyID: ak,
183+ secretAccessKey: sk,
184+ sessionToken: p["aws_session_token"],
185+ }, nil
186+}
187+
188+func (c *bedrockClient) Name() string { return "amazon-bedrock" }
189+
190+// ---- request building ----
191+
192+type bedrockMessage struct {
193+ Role string `json:"role"`
194+ Content []map[string]interface{} `json:"content"`
195+}
196+
197+type bedrockToolSpec struct {
198+ ToolSpec struct {
199+ Name string `json:"name"`
200+ Description string `json:"description,omitempty"`
201+ InputSchema struct {
202+ JSON json.RawMessage `json:"json"`
203+ } `json:"inputSchema"`
204+ } `json:"toolSpec"`
205+}
206+
207+type bedrockRequest struct {
208+ Messages []bedrockMessage `json:"messages"`
209+ System []map[string]string `json:"system,omitempty"`
210+ InferenceConfig struct {
211+ MaxTokens int `json:"maxTokens,omitempty"`
212+ Temperature *float32 `json:"temperature,omitempty"`
213+ } `json:"inferenceConfig,omitempty"`
214+ ToolConfig *struct {
215+ Tools []bedrockToolSpec `json:"tools"`
216+ } `json:"toolConfig,omitempty"`
217+}
218+
219+func (c *bedrockClient) buildRequest(req Request) (*bedrockRequest, error) {
220+ out := &bedrockRequest{}
221+ if req.System != "" {
222+ out.System = []map[string]string{{"text": req.System}}
223+ }
224+ out.InferenceConfig.Temperature = req.Temperature
225+ out.InferenceConfig.MaxTokens = req.MaxTokens
226+ if out.InferenceConfig.MaxTokens == 0 {
227+ out.InferenceConfig.MaxTokens = 4096
228+ }
229+ for _, m := range req.Messages {
230+ role := string(m.Role)
231+ if role == "tool" {
232+ role = "user"
233+ }
234+ bm := bedrockMessage{Role: role}
235+ for _, c := range m.Content {
236+ switch v := c.(type) {
237+ case TextBlock:
238+ bm.Content = append(bm.Content, map[string]interface{}{"text": v.Text})
239+ case ToolCallBlock:
240+ var input map[string]interface{}
241+ _ = json.Unmarshal(v.Arguments, &input)
242+ if input == nil {
243+ input = map[string]interface{}{}
244+ }
245+ bm.Content = append(bm.Content, map[string]interface{}{
246+ "toolUse": map[string]interface{}{
247+ "toolUseId": v.ID, "name": v.Name, "input": input,
248+ },
249+ })
250+ case ToolResultBlock:
251+ var resultContent []map[string]interface{}
252+ for _, inner := range v.Content {
253+ if tb, ok := inner.(TextBlock); ok {
254+ resultContent = append(resultContent, map[string]interface{}{"text": tb.Text})
255+ }
256+ }
257+ status := "success"
258+ if v.IsError {
259+ status = "error"
260+ }
261+ bm.Content = append(bm.Content, map[string]interface{}{
262+ "toolResult": map[string]interface{}{
263+ "toolUseId": v.CallID,
264+ "content": resultContent,
265+ "status": status,
266+ },
267+ })
268+ }
269+ }
270+ out.Messages = append(out.Messages, bm)
271+ }
272+ if len(req.Tools) > 0 {
273+ tc := struct {
274+ Tools []bedrockToolSpec `json:"tools"`
275+ }{}
276+ for _, t := range req.Tools {
277+ var ts bedrockToolSpec
278+ ts.ToolSpec.Name = t.Name
279+ ts.ToolSpec.Description = t.Description
280+ ts.ToolSpec.InputSchema.JSON = t.Schema
281+ tc.Tools = append(tc.Tools, ts)
282+ }
283+ out.ToolConfig = &tc
284+ }
285+ return out, nil
286+}
287+
288+func (c *bedrockClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
289+ wire, err := c.buildRequest(req)
290+ if err != nil {
291+ return nil, err
292+ }
293+ body, err := json.Marshal(wire)
294+ if err != nil {
295+ return nil, err
296+ }
297+ url := c.baseURL + "/model/" + req.Model + "/converse-stream"
298+ httpReq, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(body))
299+ if err != nil {
300+ return nil, err
301+ }
302+ httpReq.Header.Set("content-type", "application/json")
303+ httpReq.Header.Set("accept", "application/vnd.amazon.eventstream")
304+ if c.bearerToken != "" {
305+ httpReq.Header.Set("authorization", "Bearer "+c.bearerToken)
306+ } else if c.sigv4 != nil {
307+ if err := signSigV4(httpReq, body, "bedrock", c.region, c.sigv4, time.Now().UTC()); err != nil {
308+ return nil, fmt.Errorf("bedrock: sigv4 sign: %w", err)
309+ }
310+ } else {
311+ return nil, fmt.Errorf("bedrock: no auth configured")
312+ }
313+ resp, err := c.http.Do(httpReq)
314+ if err != nil {
315+ return nil, fmt.Errorf("bedrock: %w", err)
316+ }
317+ if resp.StatusCode != http.StatusOK {
318+ b, _ := io.ReadAll(resp.Body)
319+ resp.Body.Close()
320+ return nil, fmt.Errorf("bedrock: http %d: %s", resp.StatusCode, strings.TrimSpace(string(b)))
321+ }
322+ out := make(chan Event, 16)
323+ go c.runStream(ctx, resp, req, out)
324+ return out, nil
325+}
326+
327+func (c *bedrockClient) runStream(ctx context.Context, resp *http.Response, req Request, out chan<- Event) {
328+ defer close(out)
329+ defer resp.Body.Close()
330+
331+ out <- EventStart{Provider: "amazon-bedrock", Model: req.Model}
332+
333+ // State accumulated across deltas.
334+ contentBlocks := map[int]*bedrockBlockState{}
335+ stop := StopEnd
336+ finalMsg := Message{Role: RoleAssistant, Time: time.Now()}
337+ var usage Usage
338+
339+ for {
340+ if ctx.Err() != nil {
341+ stop = StopAborted
342+ break
343+ }
344+ evt, err := readEventStreamMessage(resp.Body)
345+ if err != nil {
346+ if err == io.EOF {
347+ break
348+ }
349+ out <- EventDone{Stop: StopError, Err: err, Message: finalMsg}
350+ return
351+ }
352+ eventType := evt.headerString(":event-type")
353+ messageType := evt.headerString(":message-type")
354+ if messageType == "exception" {
355+ out <- EventDone{Stop: StopError, Err: fmt.Errorf("bedrock exception (%s): %s", evt.headerString(":exception-type"), string(evt.payload)), Message: finalMsg}
356+ return
357+ }
358+ switch eventType {
359+ case "messageStart":
360+ // nothing to do; role is always "assistant"
361+ case "contentBlockStart":
362+ var d struct {
363+ ContentBlockIndex int `json:"contentBlockIndex"`
364+ Start struct {
365+ ToolUse *struct {
366+ ToolUseID string `json:"toolUseId"`
367+ Name string `json:"name"`
368+ } `json:"toolUse"`
369+ } `json:"start"`
370+ }
371+ if err := json.Unmarshal(evt.payload, &d); err != nil {
372+ continue
373+ }
374+ st := &bedrockBlockState{}
375+ contentBlocks[d.ContentBlockIndex] = st
376+ if d.Start.ToolUse != nil {
377+ st.isToolUse = true
378+ st.toolID = d.Start.ToolUse.ToolUseID
379+ st.toolName = d.Start.ToolUse.Name
380+ out <- EventToolStart{ID: st.toolID, Name: st.toolName}
381+ }
382+ case "contentBlockDelta":
383+ var d struct {
384+ ContentBlockIndex int `json:"contentBlockIndex"`
385+ Delta struct {
386+ Text string `json:"text"`
387+ ToolUse *struct {
388+ Input string `json:"input"`
389+ } `json:"toolUse"`
390+ } `json:"delta"`
391+ }
392+ if err := json.Unmarshal(evt.payload, &d); err != nil {
393+ continue
394+ }
395+ st := contentBlocks[d.ContentBlockIndex]
396+ if st == nil {
397+ continue
398+ }
399+ if d.Delta.Text != "" {
400+ st.text.WriteString(d.Delta.Text)
401+ out <- EventTextDelta{Delta: d.Delta.Text}
402+ }
403+ if d.Delta.ToolUse != nil && d.Delta.ToolUse.Input != "" {
404+ st.toolArgs.WriteString(d.Delta.ToolUse.Input)
405+ out <- EventToolArgs{ID: st.toolID, Delta: d.Delta.ToolUse.Input}
406+ }
407+ case "contentBlockStop":
408+ var d struct {
409+ ContentBlockIndex int `json:"contentBlockIndex"`
410+ }
411+ if err := json.Unmarshal(evt.payload, &d); err != nil {
412+ continue
413+ }
414+ st := contentBlocks[d.ContentBlockIndex]
415+ if st == nil {
416+ continue
417+ }
418+ if st.isToolUse {
419+ args := st.toolArgs.String()
420+ if args == "" {
421+ args = "{}"
422+ }
423+ finalMsg.Content = append(finalMsg.Content, ToolCallBlock{
424+ ID: st.toolID, Name: st.toolName, Arguments: json.RawMessage(args),
425+ })
426+ out <- EventToolEnd{ID: st.toolID}
427+ } else if st.text.Len() > 0 {
428+ finalMsg.Content = append(finalMsg.Content, TextBlock{Text: st.text.String()})
429+ }
430+ case "messageStop":
431+ var d struct {
432+ StopReason string `json:"stopReason"`
433+ }
434+ _ = json.Unmarshal(evt.payload, &d)
435+ switch d.StopReason {
436+ case "tool_use":
437+ stop = StopToolUse
438+ case "end_turn":
439+ stop = StopEnd
440+ case "max_tokens":
441+ stop = StopLength
442+ default:
443+ stop = StopEnd
444+ }
445+ case "metadata":
446+ var d struct {
447+ Usage struct {
448+ InputTokens int `json:"inputTokens"`
449+ OutputTokens int `json:"outputTokens"`
450+ CacheReadInputTokens int `json:"cacheReadInputTokens"`
451+ CacheWriteInputTokens int `json:"cacheWriteInputTokens"`
452+ } `json:"usage"`
453+ }
454+ if err := json.Unmarshal(evt.payload, &d); err == nil {
455+ usage.InputTokens = d.Usage.InputTokens
456+ usage.OutputTokens = d.Usage.OutputTokens
457+ usage.CacheReadTokens = d.Usage.CacheReadInputTokens
458+ usage.CacheWriteTokens = d.Usage.CacheWriteInputTokens
459+ if m, err := FindModel("amazon-bedrock", req.Model); err == nil {
460+ usage.CostUSD = ComputeCost(m, usage)
461+ }
462+ out <- EventUsage{Usage: usage}
463+ }
464+ }
465+ }
466+ out <- EventDone{Stop: stop, Message: finalMsg}
467+}
468+
469+type bedrockBlockState struct {
470+ isToolUse bool
471+ toolID string
472+ toolName string
473+ toolArgs strings.Builder
474+ text strings.Builder
475+}
476+
477+// ---- event-stream binary framing parser ----
478+
479+type eventStreamMessage struct {
480+ headers map[string]string
481+ payload []byte
482+}
483+
484+func (m *eventStreamMessage) headerString(name string) string { return m.headers[name] }
485+
486+// readEventStreamMessage reads one event-stream message from r. Returns
487+// io.EOF when the stream is finished cleanly.
488+func readEventStreamMessage(r io.Reader) (*eventStreamMessage, error) {
489+ var prelude [8]byte
490+ if _, err := io.ReadFull(r, prelude[:]); err != nil {
491+ return nil, err
492+ }
493+ totalLen := binary.BigEndian.Uint32(prelude[0:4])
494+ headersLen := binary.BigEndian.Uint32(prelude[4:8])
495+ if totalLen < 16 || headersLen > totalLen-16 {
496+ return nil, fmt.Errorf("bedrock: malformed event-stream prelude (total=%d headers=%d)", totalLen, headersLen)
497+ }
498+ // Skip prelude CRC (4 bytes); we don't validate.
499+ var preludeCRC [4]byte
500+ if _, err := io.ReadFull(r, preludeCRC[:]); err != nil {
501+ return nil, err
502+ }
503+ headersBuf := make([]byte, headersLen)
504+ if _, err := io.ReadFull(r, headersBuf); err != nil {
505+ return nil, err
506+ }
507+ payloadLen := int(totalLen) - 16 - int(headersLen)
508+ payload := make([]byte, payloadLen)
509+ if _, err := io.ReadFull(r, payload); err != nil {
510+ return nil, err
511+ }
512+ var trailer [4]byte
513+ if _, err := io.ReadFull(r, trailer[:]); err != nil {
514+ return nil, err
515+ }
516+ // Parse headers.
517+ hdrs := map[string]string{}
518+ i := 0
519+ for i < len(headersBuf) {
520+ if i+1 > len(headersBuf) {
521+ break
522+ }
523+ nameLen := int(headersBuf[i])
524+ i++
525+ if i+nameLen > len(headersBuf) {
526+ return nil, fmt.Errorf("bedrock: header name overflow")
527+ }
528+ name := string(headersBuf[i : i+nameLen])
529+ i += nameLen
530+ if i+1 > len(headersBuf) {
531+ return nil, fmt.Errorf("bedrock: header value-type missing")
532+ }
533+ valueType := headersBuf[i]
534+ i++
535+ // Type 7 = string with 2-byte length. Type 6 = byte-array similar.
536+ // Other types (bool/int/long/timestamp/uuid) we treat as opaque
537+ // strings since bedrock's event-type headers are always type 7.
538+ switch valueType {
539+ case 7, 6:
540+ if i+2 > len(headersBuf) {
541+ return nil, fmt.Errorf("bedrock: header value length missing")
542+ }
543+ vLen := int(binary.BigEndian.Uint16(headersBuf[i : i+2]))
544+ i += 2
545+ if i+vLen > len(headersBuf) {
546+ return nil, fmt.Errorf("bedrock: header value overflow")
547+ }
548+ hdrs[name] = string(headersBuf[i : i+vLen])
549+ i += vLen
550+ case 0: // bool true
551+ hdrs[name] = "true"
552+ case 1: // bool false
553+ hdrs[name] = "false"
554+ case 2: // byte
555+ if i+1 > len(headersBuf) {
556+ return nil, fmt.Errorf("bedrock: header byte missing")
557+ }
558+ i++
559+ case 3: // int16
560+ if i+2 > len(headersBuf) {
561+ return nil, fmt.Errorf("bedrock: header int16 missing")
562+ }
563+ i += 2
564+ case 4: // int32
565+ if i+4 > len(headersBuf) {
566+ return nil, fmt.Errorf("bedrock: header int32 missing")
567+ }
568+ i += 4
569+ case 5: // int64
570+ if i+8 > len(headersBuf) {
571+ return nil, fmt.Errorf("bedrock: header int64 missing")
572+ }
573+ i += 8
574+ case 8: // timestamp (int64)
575+ if i+8 > len(headersBuf) {
576+ return nil, fmt.Errorf("bedrock: header timestamp missing")
577+ }
578+ i += 8
579+ case 9: // uuid (16 bytes)
580+ if i+16 > len(headersBuf) {
581+ return nil, fmt.Errorf("bedrock: header uuid missing")
582+ }
583+ i += 16
584+ default:
585+ return nil, fmt.Errorf("bedrock: unknown header value-type %d", valueType)
586+ }
587+ }
588+ return &eventStreamMessage{headers: hdrs, payload: payload}, nil
589+}
590+
591+// ---- SigV4 signing ----
592+//
593+// AWS Signature Version 4. Implemented from scratch so we don't pull in
594+// aws-sdk-go-v2 (large transitive dep tree just for one signer). The
595+// algorithm is documented at:
596+// https://docs.aws.amazon.com/general/latest/gr/sigv4-signed-request-examples.html
597+//
598+// Steps:
599+//
600+// 1. Build the canonical request: method + path + canonical query +
601+// canonical headers + signed headers + payload-hash.
602+// 2. Build the string-to-sign: algorithm + amz-date + credential scope +
603+// sha256(canonical request).
604+// 3. Derive the signing key by chaining HMAC-SHA256:
605+// kSecret -> kDate -> kRegion -> kService -> kSigning
606+// 4. signature = HMAC-SHA256(kSigning, stringToSign)
607+// 5. Authorization header: AWS4-HMAC-SHA256 Credential=..., SignedHeaders=..., Signature=...
608+//
609+// Payload hash is sha256 of the JSON body. For Bedrock's streaming
610+// endpoint this works because the request body is fully buffered before
611+// the call — we don't need the streaming-payload variant.
612+
613+func sigHMAC(key, msg []byte) []byte {
614+ m := hmac.New(sha256.New, key)
615+ m.Write(msg)
616+ return m.Sum(nil)
617+}
618+
619+func sigSHA256Hex(b []byte) string {
620+ h := sha256.Sum256(b)
621+ return hex.EncodeToString(h[:])
622+}
623+
624+func signSigV4(req *http.Request, payload []byte, service, region string, creds *bedrockSigV4Creds, now time.Time) error {
625+ amzDate := now.Format("20060102T150405Z")
626+ shortDate := now.Format("20060102")
627+ credentialScope := shortDate + "/" + region + "/" + service + "/aws4_request"
628+
629+ // Required headers.
630+ req.Header.Set("host", req.URL.Host)
631+ req.Header.Set("x-amz-date", amzDate)
632+ if creds.sessionToken != "" {
633+ req.Header.Set("x-amz-security-token", creds.sessionToken)
634+ }
635+ payloadHash := sigSHA256Hex(payload)
636+ req.Header.Set("x-amz-content-sha256", payloadHash)
637+
638+ // Canonical headers: lowercase name, trimmed value, sorted by name,
639+ // each on its own line ending with \n.
640+ var names []string
641+ values := map[string]string{}
642+ for k, vs := range req.Header {
643+ lower := strings.ToLower(k)
644+ // Skip authorization (we're computing it) and any
645+ // connection-specific headers.
646+ if lower == "authorization" {
647+ continue
648+ }
649+ names = append(names, lower)
650+ values[lower] = strings.Join(vs, ",")
651+ }
652+ // host is special — it's not in req.Header by default; set above.
653+ sort.Strings(names)
654+ var canonHdrs strings.Builder
655+ for _, n := range names {
656+ canonHdrs.WriteString(n)
657+ canonHdrs.WriteByte(':')
658+ canonHdrs.WriteString(strings.TrimSpace(values[n]))
659+ canonHdrs.WriteByte('\n')
660+ }
661+ signedHeaders := strings.Join(names, ";")
662+
663+ // Canonical query string: keys sorted by name, each key and value
664+ // URI-escaped (RFC 3986).
665+ q := req.URL.Query()
666+ var qKeys []string
667+ for k := range q {
668+ qKeys = append(qKeys, k)
669+ }
670+ sort.Strings(qKeys)
671+ var canonQuery strings.Builder
672+ for i, k := range qKeys {
673+ if i > 0 {
674+ canonQuery.WriteByte('&')
675+ }
676+ canonQuery.WriteString(url.QueryEscape(k))
677+ canonQuery.WriteByte('=')
678+ canonQuery.WriteString(url.QueryEscape(q.Get(k)))
679+ }
680+
681+ // Canonical request.
682+ canonReq := req.Method + "\n" +
683+ canonicalSigV4Path(req.URL.Path) + "\n" +
684+ canonQuery.String() + "\n" +
685+ canonHdrs.String() + "\n" +
686+ signedHeaders + "\n" +
687+ payloadHash
688+
689+ // String to sign.
690+ stringToSign := "AWS4-HMAC-SHA256\n" + amzDate + "\n" + credentialScope + "\n" + sigSHA256Hex([]byte(canonReq))
691+
692+ // Signing key.
693+ kDate := sigHMAC([]byte("AWS4"+creds.secretAccessKey), []byte(shortDate))
694+ kRegion := sigHMAC(kDate, []byte(region))
695+ kService := sigHMAC(kRegion, []byte(service))
696+ kSigning := sigHMAC(kService, []byte("aws4_request"))
697+
698+ // Signature.
699+ signature := hex.EncodeToString(sigHMAC(kSigning, []byte(stringToSign)))
700+
701+ // Authorization header.
702+ auth := "AWS4-HMAC-SHA256 " +
703+ "Credential=" + creds.accessKeyID + "/" + credentialScope + ", " +
704+ "SignedHeaders=" + signedHeaders + ", " +
705+ "Signature=" + signature
706+ req.Header.Set("Authorization", auth)
707+ return nil
708+}
709+
710+// canonicalSigV4Path returns the path with each segment URI-encoded
711+// according to SigV4 rules (everything except unreserved chars
712+// A-Za-z0-9-_.~ and / between segments). Bedrock model IDs contain
713+// dots, colons, and slashes; the colons must be percent-encoded.
714+func canonicalSigV4Path(p string) string {
715+ if p == "" {
716+ return "/"
717+ }
718+ segments := strings.Split(p, "/")
719+ for i, s := range segments {
720+ segments[i] = sigV4EncodeSegment(s)
721+ }
722+ return strings.Join(segments, "/")
723+}
724+
725+func sigV4EncodeSegment(s string) string {
726+ var b strings.Builder
727+ for i := 0; i < len(s); i++ {
728+ c := s[i]
729+ // Unreserved per RFC 3986.
730+ if (c >= 'A' && c <= 'Z') || (c >= 'a' && c <= 'z') || (c >= '0' && c <= '9') ||
731+ c == '-' || c == '_' || c == '.' || c == '~' {
732+ b.WriteByte(c)
733+ continue
734+ }
735+ fmt.Fprintf(&b, "%%%02X", c)
736+ }
737+ return b.String()
738+}
A internal/provider/amazon_bedrock_test.go +83 -0
@@ -0,0 +1,83 @@
1+package provider
2+
3+import (
4+ "bytes"
5+ "net/http"
6+ "strings"
7+ "testing"
8+ "time"
9+)
10+
11+// TestSigV4Known checks our signer against AWS's published test vector
12+// from the SigV4 documentation. The expected Authorization header below
13+// is taken from "get-vanilla" in the suite (adapted: we use POST with
14+// an empty body since Bedrock is always POST).
15+//
16+// We don't run the official suite (it lives in S3 as a tarball and the
17+// signing rules vary slightly between examples); instead this test
18+// pins a fixed input/output so future refactors can't silently break
19+// the signer.
20+func TestSigV4Deterministic(t *testing.T) {
21+ creds := &bedrockSigV4Creds{
22+ accessKeyID: "AKIDEXAMPLE",
23+ secretAccessKey: "wJalrXUtnFEMI/K7MDENG+bPxRfiCYEXAMPLEKEY",
24+ }
25+ body := []byte(`{"hello":"world"}`)
26+ req, err := http.NewRequest("POST", "https://bedrock-runtime.us-east-1.amazonaws.com/model/foo/converse-stream", bytes.NewReader(body))
27+ if err != nil {
28+ t.Fatal(err)
29+ }
30+ req.Header.Set("content-type", "application/json")
31+ when := time.Date(2024, 1, 15, 12, 0, 0, 0, time.UTC)
32+ if err := signSigV4(req, body, "bedrock", "us-east-1", creds, when); err != nil {
33+ t.Fatal(err)
34+ }
35+ auth := req.Header.Get("Authorization")
36+ if !strings.HasPrefix(auth, "AWS4-HMAC-SHA256 ") {
37+ t.Fatalf("missing AWS4 prefix: %q", auth)
38+ }
39+ if !strings.Contains(auth, "Credential=AKIDEXAMPLE/20240115/us-east-1/bedrock/aws4_request") {
40+ t.Errorf("wrong credential scope: %q", auth)
41+ }
42+ if !strings.Contains(auth, "SignedHeaders=") {
43+ t.Errorf("missing SignedHeaders: %q", auth)
44+ }
45+ if !strings.Contains(auth, "Signature=") {
46+ t.Errorf("missing Signature: %q", auth)
47+ }
48+ // Sign again with identical inputs -> identical signature (determinism).
49+ req2, _ := http.NewRequest("POST", "https://bedrock-runtime.us-east-1.amazonaws.com/model/foo/converse-stream", bytes.NewReader(body))
50+ req2.Header.Set("content-type", "application/json")
51+ if err := signSigV4(req2, body, "bedrock", "us-east-1", creds, when); err != nil {
52+ t.Fatal(err)
53+ }
54+ if req2.Header.Get("Authorization") != auth {
55+ t.Errorf("signature not deterministic:\n a=%q\n b=%q", auth, req2.Header.Get("Authorization"))
56+ }
57+}
58+
59+func TestSigV4PathEncoding(t *testing.T) {
60+ // Bedrock model IDs include colons and dots. The colon must encode
61+ // to %3A in the canonical path (per SigV4 rules).
62+ cases := map[string]string{
63+ "/model/anthropic.claude-sonnet-4-5-20250929-v1:0/converse-stream": "/model/anthropic.claude-sonnet-4-5-20250929-v1%3A0/converse-stream",
64+ "/": "/",
65+ "": "/",
66+ "/a": "/a",
67+ }
68+ for in, want := range cases {
69+ got := canonicalSigV4Path(in)
70+ if got != want {
71+ t.Errorf("canonicalSigV4Path(%q) = %q, want %q", in, got, want)
72+ }
73+ }
74+}
75+
76+func TestReadAWSCredentialsFile(t *testing.T) {
77+ // Smoke test: parses an INI-ish file when present. We don't require
78+ // the file to exist on the test host.
79+ if _, err := readAWSCredentialsFile("default"); err != nil {
80+ // missing file or missing profile is fine — both are non-panics
81+ t.Logf("no aws creds available (expected on CI): %v", err)
82+ }
83+}
M internal/provider/anthropic.go +57 -5
@@ -60,6 +60,15 @@ type anthropicClient struct {
6060 baseURL string
6161 oauthTok string // when non-empty, send Bearer auth instead of x-api-key
6262 http *http.Client
63+
64+ // name overrides the default "anthropic" identity. Anthropic-Messages-
65+ // compatible third-party endpoints (kimi-coding, fireworks, minimax,
66+ // vercel-ai-gateway, etc.) set this so cost lookup / logging / image-
67+ // stripping logic can route on a stable provider id.
68+ name string
69+
70+ // headers carries extra request headers (e.g. Kimi Code's X-Msh-*).
71+ headers map[string]string
6372 }
6473
6574 // NewAnthropic creates an Anthropic client using an API key. baseURL may be empty.
@@ -86,7 +95,12 @@ func NewAnthropicOAuth(accessToken, baseURL string) Client {
8695 }
8796 }
8897
89-func (c *anthropicClient) Name() string { return "anthropic" }
98+func (c *anthropicClient) Name() string {
99+ if c.name != "" {
100+ return c.name
101+ }
102+ return "anthropic"
103+}
90104
91105 // ---- wire types ----
92106
@@ -166,7 +180,18 @@ type anthRequest struct {
166180 // ---- request building ----
167181
168182 func (c *anthropicClient) buildRequest(req Request) (*anthRequest, error) {
169- m, err := FindModel("anthropic", req.Model)
183+ // Look up under the client's actual provider id, which may be "anthropic"
184+ // (default) or a third party that speaks the Anthropic Messages API
185+ // (kimi, fireworks, minimax, vercel-ai-gateway, ...). Falling back to a
186+ // provider-agnostic lookup keeps things working for catalog-less
187+ // configurations (e.g. user passes --model on an obscure third party).
188+ m, err := FindModel(c.Name(), req.Model)
189+ if err != nil {
190+ if m2, err2 := FindModel("", req.Model); err2 == nil {
191+ m = m2
192+ err = nil
193+ }
194+ }
170195 if err != nil {
171196 return nil, err
172197 }
@@ -250,6 +275,11 @@ func (c *anthropicClient) buildRequest(req Request) (*anthRequest, error) {
250275
251276 // Convert messages. Anthropic's wire format has only "user" and
252277 // "assistant" roles; tool_result blocks live inside user messages.
278+ // Some image-capable providers (for example Gemini image generation)
279+ // can emit assistant image blocks. Anthropic only accepts image blocks
280+ // in user messages, so strip assistant-side images when switching
281+ // providers mid-session. The saved-path TextBlock emitted beside the
282+ // image keeps the artifact reachable in chat.
253283 //
254284 // CRITICAL: tool_result blocks go into their OWN new user
255285 // message, they are NOT merged into the preceding user message.
@@ -279,7 +309,7 @@ func (c *anthropicClient) buildRequest(req Request) (*anthRequest, error) {
279309 case RoleAssistant:
280310 out.Messages = append(out.Messages, anthMessage{
281311 Role: "assistant",
282- Content: convertAnthContent(msg.Content, renameTools),
312+ Content: convertAnthContent(filterAnthAssistantContent(msg.Content), renameTools),
283313 })
284314 }
285315 }
@@ -343,6 +373,17 @@ func anthropicReasoningBudget(level string) int {
343373 }
344374 }
345375
376+func filterAnthAssistantContent(blocks []Content) []Content {
377+ filtered := make([]Content, 0, len(blocks))
378+ for _, b := range blocks {
379+ if _, ok := b.(ImageBlock); ok {
380+ continue
381+ }
382+ filtered = append(filtered, b)
383+ }
384+ return filtered
385+}
386+
346387 func convertAnthContent(blocks []Content, renameTools bool) []interface{} {
347388 out := make([]interface{}, 0, len(blocks))
348389 for _, b := range blocks {
@@ -475,6 +516,12 @@ func (c *anthropicClient) Stream(ctx context.Context, req Request) (<-chan Event
475516 httpReq.Header.Set("accept", "text/event-stream")
476517 httpReq.Header.Set("x-api-key", c.apiKey)
477518 }
519+ // Extra headers (set by anthropic-messages-compatible third parties
520+ // — kimi-coding's X-Msh-*, copilot's Editor-Plugin-Version, etc.).
521+ // Applied last so callers can override defaults if they really need to.
522+ for k, v := range c.headers {
523+ httpReq.Header.Set(k, v)
524+ }
478525 return httpReq, nil
479526 }
480527
@@ -497,8 +544,13 @@ func (c *anthropicClient) runStream(ctx context.Context, resp *http.Response, re
497544 defer close(out)
498545 defer resp.Body.Close()
499546
500- model, _ := FindModel("anthropic", req.Model)
501- out <- EventStart{Model: req.Model, Provider: "anthropic"}
547+ // Same lookup-by-actual-provider-id pattern as buildRequest, so cost
548+ // calculation works for third-party Anthropic-Messages endpoints.
549+ model, _ := FindModel(c.Name(), req.Model)
550+ if model.ID == "" {
551+ model, _ = FindModel("", req.Model)
552+ }
553+ out <- EventStart{Model: req.Model, Provider: c.Name()}
502554
503555 raw := make(chan sseEvent, 16)
504556 go readSSE(resp.Body, raw)
A internal/provider/azure_openai.go +135 -0
@@ -0,0 +1,135 @@
1+package provider
2+
3+// Azure OpenAI client.
4+//
5+// Strategy: reuse the existing openaiClient (Chat Completions wire format)
6+// but with three Azure-specific adjustments handled at construction time:
7+//
8+// - URL shape: `{base}/openai/deployments/{model}/chat/completions?api-version={v}`
9+// This is wired by rewriting the model id into a path-prefix so the
10+// downstream code that appends `/v1/chat/completions` lands on the
11+// right endpoint. We achieve this by routing requests through a small
12+// RoundTripper that rewrites the outgoing URL before the request hits
13+// the network. The `openaiClient` itself is untouched.
14+// - Auth: `api-key: <key>` header instead of `Authorization: Bearer ...`.
15+// Sent via openaiClient.headers; the standard Authorization header
16+// remains but Azure ignores it.
17+// - API version: appended as `?api-version=...` query string by the
18+// same RoundTripper.
19+//
20+// We deliberately use Chat Completions (not the newer Responses API) so we
21+// don't have to duplicate the entire openai-responses wire client. The
22+// agent loop only needs streamed text + tool calls, both of which work
23+// fine on Azure's Chat Completions deployment.
24+
25+import (
26+ "fmt"
27+ "net/http"
28+ "net/url"
29+ "os"
30+ "strings"
31+)
32+
33+const defaultAzureAPIVersion = "2024-10-21"
34+
35+// azureRewriteTransport rewrites every outgoing OpenAI Chat Completions
36+// URL into the Azure deployment-scoped shape:
37+// /openai/v1/chat/completions ->
38+// /openai/deployments/{model}/chat/completions?api-version={v}
39+//
40+// The model id is read from the JSON body of the POST so we don't need to
41+// thread it through the client at construction time. This keeps the
42+// openaiClient request-builder untouched.
43+type azureRewriteTransport struct {
44+ inner http.RoundTripper
45+ apiVersion string
46+}
47+
48+func (t *azureRewriteTransport) RoundTrip(req *http.Request) (*http.Response, error) {
49+ if strings.Contains(req.URL.Path, "/chat/completions") {
50+ // Extract model id from the JSON body. The body is a *bytes.Reader
51+ // (set up by openaiClient.Stream) so we can peek without consuming
52+ // the request — Go's http.NewRequestWithContext gives us
53+ // GetBody() automatically for *bytes.Reader inputs.
54+ modelID := ""
55+ if req.GetBody != nil {
56+ if body, err := req.GetBody(); err == nil {
57+ buf := make([]byte, 8192)
58+ n, _ := body.Read(buf)
59+ _ = body.Close()
60+ // Cheap manual scan: look for `"model":"<id>"`.
61+ s := string(buf[:n])
62+ if i := strings.Index(s, `"model":"`); i >= 0 {
63+ rest := s[i+len(`"model":"`):]
64+ if j := strings.Index(rest, `"`); j >= 0 {
65+ modelID = rest[:j]
66+ }
67+ }
68+ }
69+ }
70+ if modelID == "" {
71+ return nil, fmt.Errorf("azure: could not extract model id from request body")
72+ }
73+ // Strip whatever prefix the openaiClient picked and replace with
74+ // Azure's deployment path. Both `/openai/v1/chat/completions` and
75+ // plain `/chat/completions` are normalised.
76+ newPath := "/openai/deployments/" + url.PathEscape(modelID) + "/chat/completions"
77+ // Anchor at the host root; the openaiClient may have used a
78+ // base URL with no path or with `/openai/v1`.
79+ req.URL.Path = newPath
80+ q := req.URL.Query()
81+ q.Set("api-version", t.apiVersion)
82+ req.URL.RawQuery = q.Encode()
83+ }
84+ return t.inner.RoundTrip(req)
85+}
86+
87+// NewAzureOpenAI returns an Azure OpenAI client.
88+//
89+// baseURL examples (any is fine; we normalise trailing slashes):
90+//
91+// https://my-resource.openai.azure.com
92+// https://my-resource.openai.azure.com/openai/v1
93+//
94+// apiKey is the Azure resource key. The api-version comes from
95+// AZURE_OPENAI_API_VERSION env var (default "2024-10-21").
96+func NewAzureOpenAI(apiKey, baseURL string) Client {
97+ if baseURL == "" {
98+ // Try env, then resource-name expansion.
99+ baseURL = os.Getenv("AZURE_OPENAI_BASE_URL")
100+ if baseURL == "" {
101+ if rn := os.Getenv("AZURE_OPENAI_RESOURCE_NAME"); rn != "" {
102+ baseURL = "https://" + rn + ".openai.azure.com"
103+ }
104+ }
105+ }
106+ if baseURL == "" {
107+ return &unimplementedClient{
108+ name: "azure-openai-responses",
109+ hint: "set AZURE_OPENAI_BASE_URL or AZURE_OPENAI_RESOURCE_NAME (or pass --base-url)",
110+ }
111+ }
112+ apiVersion := os.Getenv("AZURE_OPENAI_API_VERSION")
113+ if apiVersion == "" {
114+ apiVersion = defaultAzureAPIVersion
115+ }
116+ // Strip a trailing /openai/v1 (and /openai) since the rewrite transport
117+ // builds the full path itself.
118+ clean := strings.TrimRight(baseURL, "/")
119+ clean = strings.TrimSuffix(clean, "/openai/v1")
120+ clean = strings.TrimSuffix(clean, "/openai")
121+ httpClient := &http.Client{
122+ Transport: &azureRewriteTransport{
123+ inner: http.DefaultTransport,
124+ apiVersion: apiVersion,
125+ },
126+ Timeout: 0,
127+ }
128+ return &openaiClient{
129+ apiKey: apiKey,
130+ baseURL: clean,
131+ name: "azure-openai-responses",
132+ headers: map[string]string{"api-key": apiKey},
133+ http: httpClient,
134+ }
135+}
A internal/provider/catalog_builtin.go +550 -0
@@ -0,0 +1,550 @@
1+package provider
2+
3+// Built-in extended catalog. Hand-maintained list of model IDs,
4+// prices, and context windows for the third-party providers wired
5+// up in extra_providers.go.
6+//
7+// This file is data; the values are tracked verbatim from each
8+// vendor's own public pricing / model-list pages. They drift; users
9+// override via $ZOT_HOME/models.json (user entries win).
10+//
11+// Curated entries in models.go (anthropic / openai / openai-codex /
12+// the small seed list for kimi / deepseek / google) are not duplicated
13+// here; they take precedence on (provider, id) match.
14+
15+func init() { Catalog = append(Catalog, builtinCatalog...) }
16+
17+var builtinCatalog = []Model{
18+ // ----- amazon-bedrock -----
19+ {Provider: "amazon-bedrock", ID: "amazon.nova-2-lite-v1:0", DisplayName: "amazon.nova-2-lite-v1:0", ContextWindow: 128000, MaxOutput: 4096, Reasoning: false, PriceInput: 0.33, PriceOutput: 2.75, PriceCacheRead: 0, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
20+ {Provider: "amazon-bedrock", ID: "amazon.nova-lite-v1:0", DisplayName: "amazon.nova-lite-v1:0", ContextWindow: 300000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.06, PriceOutput: 0.24, PriceCacheRead: 0.015, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
21+ {Provider: "amazon-bedrock", ID: "amazon.nova-micro-v1:0", DisplayName: "amazon.nova-micro-v1:0", ContextWindow: 128000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.035, PriceOutput: 0.14, PriceCacheRead: 0.00875, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
22+ {Provider: "amazon-bedrock", ID: "amazon.nova-pro-v1:0", DisplayName: "amazon.nova-pro-v1:0", ContextWindow: 300000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.8, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
23+ {Provider: "amazon-bedrock", ID: "anthropic.claude-haiku-4-5-20251001-v1:0", DisplayName: "anthropic.claude-haiku-4-5-20251001-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.1, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
24+ {Provider: "amazon-bedrock", ID: "anthropic.claude-opus-4-1-20250805-v1:0", DisplayName: "anthropic.claude-opus-4-1-20250805-v1:0", ContextWindow: 200000, MaxOutput: 32000, Reasoning: true, PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
25+ {Provider: "amazon-bedrock", ID: "anthropic.claude-opus-4-5-20251101-v1:0", DisplayName: "anthropic.claude-opus-4-5-20251101-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
26+ {Provider: "amazon-bedrock", ID: "anthropic.claude-opus-4-6-v1", DisplayName: "anthropic.claude-opus-4-6-v1", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
27+ {Provider: "amazon-bedrock", ID: "anthropic.claude-opus-4-7", DisplayName: "anthropic.claude-opus-4-7", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
28+ {Provider: "amazon-bedrock", ID: "anthropic.claude-sonnet-4-5-20250929-v1:0", DisplayName: "anthropic.claude-sonnet-4-5-20250929-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
29+ {Provider: "amazon-bedrock", ID: "anthropic.claude-sonnet-4-6", DisplayName: "anthropic.claude-sonnet-4-6", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
30+ {Provider: "amazon-bedrock", ID: "au.anthropic.claude-haiku-4-5-20251001-v1:0", DisplayName: "au.anthropic.claude-haiku-4-5-20251001-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.1, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
31+ {Provider: "amazon-bedrock", ID: "au.anthropic.claude-opus-4-6-v1", DisplayName: "au.anthropic.claude-opus-4-6-v1", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 16.5, PriceOutput: 82.5, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
32+ {Provider: "amazon-bedrock", ID: "au.anthropic.claude-sonnet-4-5-20250929-v1:0", DisplayName: "au.anthropic.claude-sonnet-4-5-20250929-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
33+ {Provider: "amazon-bedrock", ID: "au.anthropic.claude-sonnet-4-6", DisplayName: "au.anthropic.claude-sonnet-4-6", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 3.3, PriceOutput: 16.5, PriceCacheRead: 0.33, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
34+ {Provider: "amazon-bedrock", ID: "deepseek.r1-v1:0", DisplayName: "deepseek.r1-v1:0", ContextWindow: 128000, MaxOutput: 32768, Reasoning: true, PriceInput: 1.35, PriceOutput: 5.4, PriceCacheRead: 0, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
35+ {Provider: "amazon-bedrock", ID: "deepseek.v3-v1:0", DisplayName: "deepseek.v3-v1:0", ContextWindow: 163840, MaxOutput: 81920, Reasoning: true, PriceInput: 0.58, PriceOutput: 1.68, PriceCacheRead: 0, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
36+ {Provider: "amazon-bedrock", ID: "deepseek.v3.2", DisplayName: "deepseek.v3.2", ContextWindow: 163840, MaxOutput: 81920, Reasoning: true, PriceInput: 0.62, PriceOutput: 1.85, PriceCacheRead: 0, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
37+ {Provider: "amazon-bedrock", ID: "eu.anthropic.claude-haiku-4-5-20251001-v1:0", DisplayName: "eu.anthropic.claude-haiku-4-5-20251001-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.1, BaseURL: "https://bedrock-runtime.eu-central-1.amazonaws.com"},
38+ {Provider: "amazon-bedrock", ID: "eu.anthropic.claude-opus-4-5-20251101-v1:0", DisplayName: "eu.anthropic.claude-opus-4-5-20251101-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.eu-central-1.amazonaws.com"},
39+ {Provider: "amazon-bedrock", ID: "eu.anthropic.claude-opus-4-6-v1", DisplayName: "eu.anthropic.claude-opus-4-6-v1", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.eu-central-1.amazonaws.com"},
40+ {Provider: "amazon-bedrock", ID: "eu.anthropic.claude-opus-4-7", DisplayName: "eu.anthropic.claude-opus-4-7", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.eu-central-1.amazonaws.com"},
41+ {Provider: "amazon-bedrock", ID: "eu.anthropic.claude-sonnet-4-5-20250929-v1:0", DisplayName: "eu.anthropic.claude-sonnet-4-5-20250929-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://bedrock-runtime.eu-central-1.amazonaws.com"},
42+ {Provider: "amazon-bedrock", ID: "eu.anthropic.claude-sonnet-4-6", DisplayName: "eu.anthropic.claude-sonnet-4-6", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://bedrock-runtime.eu-central-1.amazonaws.com"},
43+ {Provider: "amazon-bedrock", ID: "global.anthropic.claude-haiku-4-5-20251001-v1:0", DisplayName: "global.anthropic.claude-haiku-4-5-20251001-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.1, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
44+ {Provider: "amazon-bedrock", ID: "global.anthropic.claude-opus-4-5-20251101-v1:0", DisplayName: "global.anthropic.claude-opus-4-5-20251101-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
45+ {Provider: "amazon-bedrock", ID: "global.anthropic.claude-opus-4-6-v1", DisplayName: "global.anthropic.claude-opus-4-6-v1", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
46+ {Provider: "amazon-bedrock", ID: "global.anthropic.claude-opus-4-7", DisplayName: "global.anthropic.claude-opus-4-7", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
47+ {Provider: "amazon-bedrock", ID: "global.anthropic.claude-sonnet-4-5-20250929-v1:0", DisplayName: "global.anthropic.claude-sonnet-4-5-20250929-v1:0", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
48+ {Provider: "amazon-bedrock", ID: "global.anthropic.claude-sonnet-4-6", DisplayName: "global.anthropic.claude-sonnet-4-6", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://bedrock-runtime.us-east-1.amazonaws.com"},
49+ // ----- azure-openai-responses -----
50+ {Provider: "azure-openai-responses", ID: "gpt-4", DisplayName: "gpt-4", ContextWindow: 8192, MaxOutput: 8192, Reasoning: false, PriceInput: 30, PriceOutput: 60, PriceCacheRead: 0, BaseURL: ""},
51+ {Provider: "azure-openai-responses", ID: "gpt-4-turbo", DisplayName: "gpt-4-turbo", ContextWindow: 128000, MaxOutput: 4096, Reasoning: false, PriceInput: 10, PriceOutput: 30, PriceCacheRead: 0, BaseURL: ""},
52+ {Provider: "azure-openai-responses", ID: "gpt-4.1", DisplayName: "gpt-4.1", ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false, PriceInput: 2, PriceOutput: 8, PriceCacheRead: 0.5, BaseURL: ""},
53+ {Provider: "azure-openai-responses", ID: "gpt-4.1-mini", DisplayName: "gpt-4.1-mini", ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false, PriceInput: 0.4, PriceOutput: 1.6, PriceCacheRead: 0.1, BaseURL: ""},
54+ {Provider: "azure-openai-responses", ID: "gpt-4.1-nano", DisplayName: "gpt-4.1-nano", ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false, PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.03, BaseURL: ""},
55+ {Provider: "azure-openai-responses", ID: "gpt-4o", DisplayName: "gpt-4o", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 2.5, PriceOutput: 10, PriceCacheRead: 1.25, BaseURL: ""},
56+ {Provider: "azure-openai-responses", ID: "gpt-4o-2024-05-13", DisplayName: "gpt-4o-2024-05-13", ContextWindow: 128000, MaxOutput: 4096, Reasoning: false, PriceInput: 5, PriceOutput: 15, PriceCacheRead: 0, BaseURL: ""},
57+ {Provider: "azure-openai-responses", ID: "gpt-4o-2024-08-06", DisplayName: "gpt-4o-2024-08-06", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 2.5, PriceOutput: 10, PriceCacheRead: 1.25, BaseURL: ""},
58+ {Provider: "azure-openai-responses", ID: "gpt-4o-2024-11-20", DisplayName: "gpt-4o-2024-11-20", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 2.5, PriceOutput: 10, PriceCacheRead: 1.25, BaseURL: ""},
59+ {Provider: "azure-openai-responses", ID: "gpt-4o-mini", DisplayName: "gpt-4o-mini", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0.08, BaseURL: ""},
60+ {Provider: "azure-openai-responses", ID: "gpt-5", DisplayName: "gpt-5", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: ""},
61+ {Provider: "azure-openai-responses", ID: "gpt-5-chat-latest", DisplayName: "gpt-5-chat-latest", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: ""},
62+ {Provider: "azure-openai-responses", ID: "gpt-5-codex", DisplayName: "gpt-5-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: ""},
63+ {Provider: "azure-openai-responses", ID: "gpt-5-mini", DisplayName: "gpt-5-mini", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.25, PriceOutput: 2, PriceCacheRead: 0.025, BaseURL: ""},
64+ {Provider: "azure-openai-responses", ID: "gpt-5-nano", DisplayName: "gpt-5-nano", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.05, PriceOutput: 0.4, PriceCacheRead: 0.005, BaseURL: ""},
65+ {Provider: "azure-openai-responses", ID: "gpt-5-pro", DisplayName: "gpt-5-pro", ContextWindow: 400000, MaxOutput: 272000, Reasoning: true, PriceInput: 15, PriceOutput: 120, PriceCacheRead: 0, BaseURL: ""},
66+ {Provider: "azure-openai-responses", ID: "gpt-5.1", DisplayName: "gpt-5.1", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.13, BaseURL: ""},
67+ {Provider: "azure-openai-responses", ID: "gpt-5.1-chat-latest", DisplayName: "gpt-5.1-chat-latest", ContextWindow: 128000, MaxOutput: 16384, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: ""},
68+ {Provider: "azure-openai-responses", ID: "gpt-5.1-codex", DisplayName: "gpt-5.1-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: ""},
69+ {Provider: "azure-openai-responses", ID: "gpt-5.1-codex-max", DisplayName: "gpt-5.1-codex-max", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: ""},
70+ {Provider: "azure-openai-responses", ID: "gpt-5.1-codex-mini", DisplayName: "gpt-5.1-codex-mini", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.25, PriceOutput: 2, PriceCacheRead: 0.025, BaseURL: ""},
71+ {Provider: "azure-openai-responses", ID: "gpt-5.2", DisplayName: "gpt-5.2", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175, BaseURL: ""},
72+ {Provider: "azure-openai-responses", ID: "gpt-5.2-chat-latest", DisplayName: "gpt-5.2-chat-latest", ContextWindow: 128000, MaxOutput: 16384, Reasoning: true, PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175, BaseURL: ""},
73+ {Provider: "azure-openai-responses", ID: "gpt-5.2-codex", DisplayName: "gpt-5.2-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175, BaseURL: ""},
74+ {Provider: "azure-openai-responses", ID: "gpt-5.2-pro", DisplayName: "gpt-5.2-pro", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 21, PriceOutput: 168, PriceCacheRead: 0, BaseURL: ""},
75+ // ----- cerebras -----
76+ {Provider: "cerebras", ID: "gpt-oss-120b", DisplayName: "gpt-oss-120b", ContextWindow: 131072, MaxOutput: 32768, Reasoning: true, PriceInput: 0.25, PriceOutput: 0.69, PriceCacheRead: 0, BaseURL: "https://api.cerebras.ai/v1"},
77+ {Provider: "cerebras", ID: "llama3.1-8b", DisplayName: "llama3.1-8b", ContextWindow: 32000, MaxOutput: 8000, Reasoning: false, PriceInput: 0.1, PriceOutput: 0.1, PriceCacheRead: 0, BaseURL: "https://api.cerebras.ai/v1"},
78+ {Provider: "cerebras", ID: "qwen-3-235b-a22b-instruct-2507", DisplayName: "qwen-3-235b-a22b-instruct-2507", ContextWindow: 131000, MaxOutput: 32000, Reasoning: false, PriceInput: 0.6, PriceOutput: 1.2, PriceCacheRead: 0, BaseURL: "https://api.cerebras.ai/v1"},
79+ {Provider: "cerebras", ID: "zai-glm-4.7", DisplayName: "zai-glm-4.7", ContextWindow: 131072, MaxOutput: 40000, Reasoning: false, PriceInput: 2.25, PriceOutput: 2.75, PriceCacheRead: 0, BaseURL: "https://api.cerebras.ai/v1"},
80+ // ----- cloudflare-ai-gateway -----
81+ {Provider: "cloudflare-ai-gateway", ID: "claude-3-5-haiku", DisplayName: "claude-3-5-haiku", ContextWindow: 200000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.8, PriceOutput: 4, PriceCacheRead: 0.08, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
82+ {Provider: "cloudflare-ai-gateway", ID: "claude-3-haiku", DisplayName: "claude-3-haiku", ContextWindow: 200000, MaxOutput: 4096, Reasoning: false, PriceInput: 0.25, PriceOutput: 1.25, PriceCacheRead: 0.03, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
83+ {Provider: "cloudflare-ai-gateway", ID: "claude-3-opus", DisplayName: "claude-3-opus", ContextWindow: 200000, MaxOutput: 4096, Reasoning: false, PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
84+ {Provider: "cloudflare-ai-gateway", ID: "claude-3-sonnet", DisplayName: "claude-3-sonnet", ContextWindow: 200000, MaxOutput: 4096, Reasoning: false, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
85+ {Provider: "cloudflare-ai-gateway", ID: "claude-3.5-haiku", DisplayName: "claude-3.5-haiku", ContextWindow: 200000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.8, PriceOutput: 4, PriceCacheRead: 0.08, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
86+ {Provider: "cloudflare-ai-gateway", ID: "claude-3.5-sonnet", DisplayName: "claude-3.5-sonnet", ContextWindow: 200000, MaxOutput: 8192, Reasoning: false, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
87+ {Provider: "cloudflare-ai-gateway", ID: "claude-haiku-4-5", DisplayName: "claude-haiku-4-5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.1, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
88+ {Provider: "cloudflare-ai-gateway", ID: "claude-opus-4", DisplayName: "claude-opus-4", ContextWindow: 200000, MaxOutput: 32000, Reasoning: true, PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
89+ {Provider: "cloudflare-ai-gateway", ID: "claude-opus-4-1", DisplayName: "claude-opus-4-1", ContextWindow: 200000, MaxOutput: 32000, Reasoning: true, PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
90+ {Provider: "cloudflare-ai-gateway", ID: "claude-opus-4-5", DisplayName: "claude-opus-4-5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
91+ {Provider: "cloudflare-ai-gateway", ID: "claude-opus-4-6", DisplayName: "claude-opus-4-6", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
92+ {Provider: "cloudflare-ai-gateway", ID: "claude-opus-4-7", DisplayName: "claude-opus-4-7", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
93+ {Provider: "cloudflare-ai-gateway", ID: "claude-sonnet-4", DisplayName: "claude-sonnet-4", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
94+ {Provider: "cloudflare-ai-gateway", ID: "claude-sonnet-4-5", DisplayName: "claude-sonnet-4-5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
95+ {Provider: "cloudflare-ai-gateway", ID: "claude-sonnet-4-6", DisplayName: "claude-sonnet-4-6", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic"},
96+ {Provider: "cloudflare-ai-gateway", ID: "gpt-4", DisplayName: "gpt-4", ContextWindow: 8192, MaxOutput: 8192, Reasoning: false, PriceInput: 30, PriceOutput: 60, PriceCacheRead: 0, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai"},
97+ {Provider: "cloudflare-ai-gateway", ID: "gpt-4-turbo", DisplayName: "gpt-4-turbo", ContextWindow: 128000, MaxOutput: 4096, Reasoning: false, PriceInput: 10, PriceOutput: 30, PriceCacheRead: 0, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai"},
98+ {Provider: "cloudflare-ai-gateway", ID: "gpt-4o", DisplayName: "gpt-4o", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 2.5, PriceOutput: 10, PriceCacheRead: 1.25, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai"},
99+ {Provider: "cloudflare-ai-gateway", ID: "gpt-4o-mini", DisplayName: "gpt-4o-mini", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0.08, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai"},
100+ {Provider: "cloudflare-ai-gateway", ID: "gpt-5.1", DisplayName: "gpt-5.1", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.13, BaseURL: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai"},
101+ // ----- cloudflare-workers-ai -----
102+ {Provider: "cloudflare-workers-ai", ID: "@cf/google/gemma-4-26b-a4b-it", DisplayName: "@cf/google/gemma-4-26b-a4b-it", ContextWindow: 256000, MaxOutput: 16384, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
103+ {Provider: "cloudflare-workers-ai", ID: "@cf/ibm-granite/granite-4.0-h-micro", DisplayName: "@cf/ibm-granite/granite-4.0-h-micro", ContextWindow: 131000, MaxOutput: 131000, Reasoning: false, PriceInput: 0.017, PriceOutput: 0.112, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
104+ {Provider: "cloudflare-workers-ai", ID: "@cf/meta/llama-3.3-70b-instruct-fp8-fast", DisplayName: "@cf/meta/llama-3.3-70b-instruct-fp8-fast", ContextWindow: 24000, MaxOutput: 24000, Reasoning: false, PriceInput: 0.293, PriceOutput: 2.253, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
105+ {Provider: "cloudflare-workers-ai", ID: "@cf/meta/llama-4-scout-17b-16e-instruct", DisplayName: "@cf/meta/llama-4-scout-17b-16e-instruct", ContextWindow: 131000, MaxOutput: 16384, Reasoning: false, PriceInput: 0.27, PriceOutput: 0.85, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
106+ {Provider: "cloudflare-workers-ai", ID: "@cf/mistralai/mistral-small-3.1-24b-instruct", DisplayName: "@cf/mistralai/mistral-small-3.1-24b-instruct", ContextWindow: 128000, MaxOutput: 128000, Reasoning: false, PriceInput: 0.351, PriceOutput: 0.555, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
107+ {Provider: "cloudflare-workers-ai", ID: "@cf/moonshotai/kimi-k2.5", DisplayName: "@cf/moonshotai/kimi-k2.5", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
108+ {Provider: "cloudflare-workers-ai", ID: "@cf/moonshotai/kimi-k2.6", DisplayName: "@cf/moonshotai/kimi-k2.6", ContextWindow: 262144, MaxOutput: 256000, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
109+ {Provider: "cloudflare-workers-ai", ID: "@cf/nvidia/nemotron-3-120b-a12b", DisplayName: "@cf/nvidia/nemotron-3-120b-a12b", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 0.5, PriceOutput: 1.5, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
110+ {Provider: "cloudflare-workers-ai", ID: "@cf/openai/gpt-oss-120b", DisplayName: "@cf/openai/gpt-oss-120b", ContextWindow: 128000, MaxOutput: 16384, Reasoning: true, PriceInput: 0.35, PriceOutput: 0.75, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
111+ {Provider: "cloudflare-workers-ai", ID: "@cf/openai/gpt-oss-20b", DisplayName: "@cf/openai/gpt-oss-20b", ContextWindow: 128000, MaxOutput: 16384, Reasoning: true, PriceInput: 0.2, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
112+ {Provider: "cloudflare-workers-ai", ID: "@cf/qwen/qwen3-30b-a3b-fp8", DisplayName: "@cf/qwen/qwen3-30b-a3b-fp8", ContextWindow: 32768, MaxOutput: 32768, Reasoning: true, PriceInput: 0.0509, PriceOutput: 0.335, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
113+ {Provider: "cloudflare-workers-ai", ID: "@cf/zai-org/glm-4.7-flash", DisplayName: "@cf/zai-org/glm-4.7-flash", ContextWindow: 131072, MaxOutput: 131072, Reasoning: true, PriceInput: 0.0605, PriceOutput: 0.4, PriceCacheRead: 0, BaseURL: "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"},
114+ // ----- fireworks -----
115+ {Provider: "fireworks", ID: "accounts/fireworks/models/deepseek-v4-flash", DisplayName: "accounts/fireworks/models/deepseek-v4-flash", ContextWindow: 1000000, MaxOutput: 384000, Reasoning: true, PriceInput: 0.14, PriceOutput: 0.28, PriceCacheRead: 0.03, BaseURL: "https://api.fireworks.ai/inference"},
116+ {Provider: "fireworks", ID: "accounts/fireworks/models/deepseek-v4-pro", DisplayName: "accounts/fireworks/models/deepseek-v4-pro", ContextWindow: 1000000, MaxOutput: 384000, Reasoning: true, PriceInput: 1.74, PriceOutput: 3.48, PriceCacheRead: 0.145, BaseURL: "https://api.fireworks.ai/inference"},
117+ {Provider: "fireworks", ID: "accounts/fireworks/models/glm-5p1", DisplayName: "accounts/fireworks/models/glm-5p1", ContextWindow: 202800, MaxOutput: 131072, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://api.fireworks.ai/inference"},
118+ {Provider: "fireworks", ID: "accounts/fireworks/models/gpt-oss-120b", DisplayName: "accounts/fireworks/models/gpt-oss-120b", ContextWindow: 131072, MaxOutput: 32768, Reasoning: true, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0.015, BaseURL: "https://api.fireworks.ai/inference"},
119+ {Provider: "fireworks", ID: "accounts/fireworks/models/gpt-oss-20b", DisplayName: "accounts/fireworks/models/gpt-oss-20b", ContextWindow: 131072, MaxOutput: 32768, Reasoning: true, PriceInput: 0.07, PriceOutput: 0.3, PriceCacheRead: 0.035, BaseURL: "https://api.fireworks.ai/inference"},
120+ {Provider: "fireworks", ID: "accounts/fireworks/models/kimi-k2p5", DisplayName: "accounts/fireworks/models/kimi-k2p5", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://api.fireworks.ai/inference"},
121+ {Provider: "fireworks", ID: "accounts/fireworks/models/kimi-k2p6", DisplayName: "accounts/fireworks/models/kimi-k2p6", ContextWindow: 262000, MaxOutput: 262000, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://api.fireworks.ai/inference"},
122+ {Provider: "fireworks", ID: "accounts/fireworks/models/minimax-m2p5", DisplayName: "accounts/fireworks/models/minimax-m2p5", ContextWindow: 196608, MaxOutput: 196608, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.03, BaseURL: "https://api.fireworks.ai/inference"},
123+ {Provider: "fireworks", ID: "accounts/fireworks/models/minimax-m2p7", DisplayName: "accounts/fireworks/models/minimax-m2p7", ContextWindow: 196608, MaxOutput: 196608, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.06, BaseURL: "https://api.fireworks.ai/inference"},
124+ {Provider: "fireworks", ID: "accounts/fireworks/models/qwen3p6-plus", DisplayName: "accounts/fireworks/models/qwen3p6-plus", ContextWindow: 128000, MaxOutput: 8192, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://api.fireworks.ai/inference"},
125+ {Provider: "fireworks", ID: "accounts/fireworks/routers/glm-5p1-fast", DisplayName: "accounts/fireworks/routers/glm-5p1-fast", ContextWindow: 202800, MaxOutput: 131072, Reasoning: true, PriceInput: 2.8, PriceOutput: 8.8, PriceCacheRead: 0.52, BaseURL: "https://api.fireworks.ai/inference"},
126+ {Provider: "fireworks", ID: "accounts/fireworks/routers/kimi-k2p6-turbo", DisplayName: "accounts/fireworks/routers/kimi-k2p6-turbo", ContextWindow: 262000, MaxOutput: 262000, Reasoning: true, PriceInput: 2, PriceOutput: 8, PriceCacheRead: 0.3, BaseURL: "https://api.fireworks.ai/inference"},
127+ // ----- github-copilot -----
128+ {Provider: "github-copilot", ID: "claude-haiku-4.5", DisplayName: "claude-haiku-4.5", ContextWindow: 144000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
129+ {Provider: "github-copilot", ID: "claude-opus-4.5", DisplayName: "claude-opus-4.5", ContextWindow: 160000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
130+ {Provider: "github-copilot", ID: "claude-opus-4.6", DisplayName: "claude-opus-4.6", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
131+ {Provider: "github-copilot", ID: "claude-opus-4.7", DisplayName: "claude-opus-4.7", ContextWindow: 144000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
132+ {Provider: "github-copilot", ID: "claude-sonnet-4.5", DisplayName: "claude-sonnet-4.5", ContextWindow: 144000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
133+ {Provider: "github-copilot", ID: "claude-sonnet-4.6", DisplayName: "claude-sonnet-4.6", ContextWindow: 1000000, MaxOutput: 32000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
134+ {Provider: "github-copilot", ID: "gemini-2.5-pro", DisplayName: "gemini-2.5-pro", ContextWindow: 128000, MaxOutput: 64000, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
135+ {Provider: "github-copilot", ID: "gemini-3-flash-preview", DisplayName: "gemini-3-flash-preview", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
136+ {Provider: "github-copilot", ID: "gemini-3.1-pro-preview", DisplayName: "gemini-3.1-pro-preview", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
137+ {Provider: "github-copilot", ID: "gemini-3.5-flash", DisplayName: "gemini-3.5-flash", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
138+ {Provider: "github-copilot", ID: "gpt-4.1", DisplayName: "gpt-4.1", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
139+ {Provider: "github-copilot", ID: "gpt-4o", DisplayName: "gpt-4o", ContextWindow: 128000, MaxOutput: 4096, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
140+ {Provider: "github-copilot", ID: "gpt-5-mini", DisplayName: "gpt-5-mini", ContextWindow: 264000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
141+ {Provider: "github-copilot", ID: "gpt-5.2", DisplayName: "gpt-5.2", ContextWindow: 264000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
142+ {Provider: "github-copilot", ID: "gpt-5.2-codex", DisplayName: "gpt-5.2-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
143+ {Provider: "github-copilot", ID: "gpt-5.3-codex", DisplayName: "gpt-5.3-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
144+ {Provider: "github-copilot", ID: "gpt-5.4", DisplayName: "gpt-5.4", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
145+ {Provider: "github-copilot", ID: "gpt-5.4-mini", DisplayName: "gpt-5.4-mini", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
146+ {Provider: "github-copilot", ID: "gpt-5.5", DisplayName: "gpt-5.5", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
147+ {Provider: "github-copilot", ID: "grok-code-fast-1", DisplayName: "grok-code-fast-1", ContextWindow: 128000, MaxOutput: 64000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.individual.githubcopilot.com"},
148+ // ----- google -----
149+ {Provider: "google", ID: "gemini-3-flash-preview", DisplayName: "gemini-3-flash-preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.05, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
150+ {Provider: "google", ID: "gemini-3-pro-preview", DisplayName: "gemini-3-pro-preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
151+ {Provider: "google", ID: "gemini-3.1-flash-lite", DisplayName: "gemini-3.1-flash-lite", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.25, PriceOutput: 1.5, PriceCacheRead: 0.025, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
152+ {Provider: "google", ID: "gemini-3.1-flash-lite-preview", DisplayName: "gemini-3.1-flash-lite-preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.25, PriceOutput: 1.5, PriceCacheRead: 0.025, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
153+ {Provider: "google", ID: "gemini-3.1-pro-preview", DisplayName: "gemini-3.1-pro-preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
154+ {Provider: "google", ID: "gemini-3.1-pro-preview-customtools", DisplayName: "gemini-3.1-pro-preview-customtools", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
155+ {Provider: "google", ID: "gemini-3.5-flash", DisplayName: "gemini-3.5-flash", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 1.5, PriceOutput: 9, PriceCacheRead: 0.15, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
156+ {Provider: "google", ID: "gemini-flash-latest", DisplayName: "gemini-flash-latest", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.3, PriceOutput: 2.5, PriceCacheRead: 0.075, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
157+ {Provider: "google", ID: "gemini-flash-lite-latest", DisplayName: "gemini-flash-lite-latest", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.025, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
158+ {Provider: "google", ID: "gemma-4-26b-a4b-it", DisplayName: "gemma-4-26b-a4b-it", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
159+ {Provider: "google", ID: "gemma-4-31b-it", DisplayName: "gemma-4-31b-it", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://generativelanguage.googleapis.com/v1beta"},
160+ // ----- google-vertex -----
161+ {Provider: "google-vertex", ID: "gemini-1.5-flash", DisplayName: "gemini-1.5-flash", ContextWindow: 1000000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0.01875, BaseURL: "https://{location}-aiplatform.googleapis.com"},
162+ {Provider: "google-vertex", ID: "gemini-1.5-flash-8b", DisplayName: "gemini-1.5-flash-8b", ContextWindow: 1000000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.0375, PriceOutput: 0.15, PriceCacheRead: 0.01, BaseURL: "https://{location}-aiplatform.googleapis.com"},
163+ {Provider: "google-vertex", ID: "gemini-1.5-pro", DisplayName: "gemini-1.5-pro", ContextWindow: 1000000, MaxOutput: 8192, Reasoning: false, PriceInput: 1.25, PriceOutput: 5, PriceCacheRead: 0.3125, BaseURL: "https://{location}-aiplatform.googleapis.com"},
164+ {Provider: "google-vertex", ID: "gemini-2.0-flash", DisplayName: "gemini-2.0-flash", ContextWindow: 1048576, MaxOutput: 8192, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0.0375, BaseURL: "https://{location}-aiplatform.googleapis.com"},
165+ {Provider: "google-vertex", ID: "gemini-2.0-flash-lite", DisplayName: "gemini-2.0-flash-lite", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0.01875, BaseURL: "https://{location}-aiplatform.googleapis.com"},
166+ {Provider: "google-vertex", ID: "gemini-2.5-flash", DisplayName: "gemini-2.5-flash", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.3, PriceOutput: 2.5, PriceCacheRead: 0.03, BaseURL: "https://{location}-aiplatform.googleapis.com"},
167+ {Provider: "google-vertex", ID: "gemini-2.5-flash-lite", DisplayName: "gemini-2.5-flash-lite", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.01, BaseURL: "https://{location}-aiplatform.googleapis.com"},
168+ {Provider: "google-vertex", ID: "gemini-2.5-flash-lite-preview-09-2025", DisplayName: "gemini-2.5-flash-lite-preview-09-2025", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.01, BaseURL: "https://{location}-aiplatform.googleapis.com"},
169+ {Provider: "google-vertex", ID: "gemini-2.5-pro", DisplayName: "gemini-2.5-pro", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://{location}-aiplatform.googleapis.com"},
170+ {Provider: "google-vertex", ID: "gemini-3-flash-preview", DisplayName: "gemini-3-flash-preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.05, BaseURL: "https://{location}-aiplatform.googleapis.com"},
171+ {Provider: "google-vertex", ID: "gemini-3-pro-preview", DisplayName: "gemini-3-pro-preview", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://{location}-aiplatform.googleapis.com"},
172+ {Provider: "google-vertex", ID: "gemini-3.1-pro-preview", DisplayName: "gemini-3.1-pro-preview", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://{location}-aiplatform.googleapis.com"},
173+ {Provider: "google-vertex", ID: "gemini-3.1-pro-preview-customtools", DisplayName: "gemini-3.1-pro-preview-customtools", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://{location}-aiplatform.googleapis.com"},
174+ // ----- groq -----
175+ {Provider: "groq", ID: "deepseek-r1-distill-llama-70b", DisplayName: "deepseek-r1-distill-llama-70b", ContextWindow: 131072, MaxOutput: 8192, Reasoning: true, PriceInput: 0.75, PriceOutput: 0.99, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
176+ {Provider: "groq", ID: "gemma2-9b-it", DisplayName: "gemma2-9b-it", ContextWindow: 8192, MaxOutput: 8192, Reasoning: false, PriceInput: 0.2, PriceOutput: 0.2, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
177+ {Provider: "groq", ID: "groq/compound", DisplayName: "groq/compound", ContextWindow: 131072, MaxOutput: 8192, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
178+ {Provider: "groq", ID: "groq/compound-mini", DisplayName: "groq/compound-mini", ContextWindow: 131072, MaxOutput: 8192, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
179+ {Provider: "groq", ID: "llama-3.1-8b-instant", DisplayName: "llama-3.1-8b-instant", ContextWindow: 131072, MaxOutput: 131072, Reasoning: false, PriceInput: 0.05, PriceOutput: 0.08, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
180+ {Provider: "groq", ID: "llama-3.3-70b-versatile", DisplayName: "llama-3.3-70b-versatile", ContextWindow: 131072, MaxOutput: 32768, Reasoning: false, PriceInput: 0.59, PriceOutput: 0.79, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
181+ {Provider: "groq", ID: "llama3-70b-8192", DisplayName: "llama3-70b-8192", ContextWindow: 8192, MaxOutput: 8192, Reasoning: false, PriceInput: 0.59, PriceOutput: 0.79, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
182+ {Provider: "groq", ID: "llama3-8b-8192", DisplayName: "llama3-8b-8192", ContextWindow: 8192, MaxOutput: 8192, Reasoning: false, PriceInput: 0.05, PriceOutput: 0.08, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
183+ {Provider: "groq", ID: "meta-llama/llama-4-maverick-17b-128e-instruct", DisplayName: "meta-llama/llama-4-maverick-17b-128e-instruct", ContextWindow: 131072, MaxOutput: 8192, Reasoning: false, PriceInput: 0.2, PriceOutput: 0.6, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
184+ {Provider: "groq", ID: "meta-llama/llama-4-scout-17b-16e-instruct", DisplayName: "meta-llama/llama-4-scout-17b-16e-instruct", ContextWindow: 131072, MaxOutput: 8192, Reasoning: false, PriceInput: 0.11, PriceOutput: 0.34, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
185+ {Provider: "groq", ID: "mistral-saba-24b", DisplayName: "mistral-saba-24b", ContextWindow: 32768, MaxOutput: 32768, Reasoning: false, PriceInput: 0.79, PriceOutput: 0.79, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
186+ {Provider: "groq", ID: "moonshotai/kimi-k2-instruct", DisplayName: "moonshotai/kimi-k2-instruct", ContextWindow: 131072, MaxOutput: 16384, Reasoning: false, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
187+ {Provider: "groq", ID: "moonshotai/kimi-k2-instruct-0905", DisplayName: "moonshotai/kimi-k2-instruct-0905", ContextWindow: 262144, MaxOutput: 16384, Reasoning: false, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
188+ {Provider: "groq", ID: "openai/gpt-oss-120b", DisplayName: "openai/gpt-oss-120b", ContextWindow: 131072, MaxOutput: 65536, Reasoning: true, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
189+ {Provider: "groq", ID: "openai/gpt-oss-20b", DisplayName: "openai/gpt-oss-20b", ContextWindow: 131072, MaxOutput: 65536, Reasoning: true, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
190+ {Provider: "groq", ID: "openai/gpt-oss-safeguard-20b", DisplayName: "openai/gpt-oss-safeguard-20b", ContextWindow: 131072, MaxOutput: 65536, Reasoning: true, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0.037, BaseURL: "https://api.groq.com/openai/v1"},
191+ {Provider: "groq", ID: "qwen-qwq-32b", DisplayName: "qwen-qwq-32b", ContextWindow: 131072, MaxOutput: 16384, Reasoning: true, PriceInput: 0.29, PriceOutput: 0.39, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
192+ {Provider: "groq", ID: "qwen/qwen3-32b", DisplayName: "qwen/qwen3-32b", ContextWindow: 131072, MaxOutput: 40960, Reasoning: true, PriceInput: 0.29, PriceOutput: 0.59, PriceCacheRead: 0, BaseURL: "https://api.groq.com/openai/v1"},
193+ // ----- huggingface -----
194+ {Provider: "huggingface", ID: "MiniMaxAI/MiniMax-M2.1", DisplayName: "MiniMaxAI/MiniMax-M2.1", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
195+ {Provider: "huggingface", ID: "MiniMaxAI/MiniMax-M2.5", DisplayName: "MiniMaxAI/MiniMax-M2.5", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.03, BaseURL: "https://router.huggingface.co/v1"},
196+ {Provider: "huggingface", ID: "MiniMaxAI/MiniMax-M2.7", DisplayName: "MiniMaxAI/MiniMax-M2.7", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.06, BaseURL: "https://router.huggingface.co/v1"},
197+ {Provider: "huggingface", ID: "Qwen/Qwen3-235B-A22B-Thinking-2507", DisplayName: "Qwen/Qwen3-235B-A22B-Thinking-2507", ContextWindow: 262144, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 3, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
198+ {Provider: "huggingface", ID: "Qwen/Qwen3-Coder-480B-A35B-Instruct", DisplayName: "Qwen/Qwen3-Coder-480B-A35B-Instruct", ContextWindow: 262144, MaxOutput: 66536, Reasoning: false, PriceInput: 2, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
199+ {Provider: "huggingface", ID: "Qwen/Qwen3-Coder-Next", DisplayName: "Qwen/Qwen3-Coder-Next", ContextWindow: 262144, MaxOutput: 65536, Reasoning: false, PriceInput: 0.2, PriceOutput: 1.5, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
200+ {Provider: "huggingface", ID: "Qwen/Qwen3-Next-80B-A3B-Instruct", DisplayName: "Qwen/Qwen3-Next-80B-A3B-Instruct", ContextWindow: 262144, MaxOutput: 66536, Reasoning: false, PriceInput: 0.25, PriceOutput: 1, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
201+ {Provider: "huggingface", ID: "Qwen/Qwen3-Next-80B-A3B-Thinking", DisplayName: "Qwen/Qwen3-Next-80B-A3B-Thinking", ContextWindow: 262144, MaxOutput: 131072, Reasoning: false, PriceInput: 0.3, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
202+ {Provider: "huggingface", ID: "Qwen/Qwen3.5-397B-A17B", DisplayName: "Qwen/Qwen3.5-397B-A17B", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0.6, PriceOutput: 3.6, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
203+ {Provider: "huggingface", ID: "XiaomiMiMo/MiMo-V2-Flash", DisplayName: "XiaomiMiMo/MiMo-V2-Flash", ContextWindow: 262144, MaxOutput: 4096, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
204+ {Provider: "huggingface", ID: "deepseek-ai/DeepSeek-R1-0528", DisplayName: "deepseek-ai/DeepSeek-R1-0528", ContextWindow: 163840, MaxOutput: 163840, Reasoning: true, PriceInput: 3, PriceOutput: 5, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
205+ {Provider: "huggingface", ID: "deepseek-ai/DeepSeek-V3.2", DisplayName: "deepseek-ai/DeepSeek-V3.2", ContextWindow: 163840, MaxOutput: 65536, Reasoning: true, PriceInput: 0.28, PriceOutput: 0.4, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
206+ {Provider: "huggingface", ID: "deepseek-ai/DeepSeek-V4-Pro", DisplayName: "deepseek-ai/DeepSeek-V4-Pro", ContextWindow: 1048576, MaxOutput: 393216, Reasoning: true, PriceInput: 1.74, PriceOutput: 3.48, PriceCacheRead: 0.145, BaseURL: "https://router.huggingface.co/v1"},
207+ {Provider: "huggingface", ID: "moonshotai/Kimi-K2-Instruct", DisplayName: "moonshotai/Kimi-K2-Instruct", ContextWindow: 131072, MaxOutput: 16384, Reasoning: false, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
208+ {Provider: "huggingface", ID: "moonshotai/Kimi-K2-Instruct-0905", DisplayName: "moonshotai/Kimi-K2-Instruct-0905", ContextWindow: 262144, MaxOutput: 16384, Reasoning: false, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
209+ {Provider: "huggingface", ID: "moonshotai/Kimi-K2-Thinking", DisplayName: "moonshotai/Kimi-K2-Thinking", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.5, PriceCacheRead: 0.15, BaseURL: "https://router.huggingface.co/v1"},
210+ {Provider: "huggingface", ID: "moonshotai/Kimi-K2.5", DisplayName: "moonshotai/Kimi-K2.5", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://router.huggingface.co/v1"},
211+ {Provider: "huggingface", ID: "moonshotai/Kimi-K2.6", DisplayName: "moonshotai/Kimi-K2.6", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://router.huggingface.co/v1"},
212+ {Provider: "huggingface", ID: "zai-org/GLM-4.7", DisplayName: "zai-org/GLM-4.7", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.2, PriceCacheRead: 0.11, BaseURL: "https://router.huggingface.co/v1"},
213+ {Provider: "huggingface", ID: "zai-org/GLM-4.7-Flash", DisplayName: "zai-org/GLM-4.7-Flash", ContextWindow: 200000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://router.huggingface.co/v1"},
214+ {Provider: "huggingface", ID: "zai-org/GLM-5", DisplayName: "zai-org/GLM-5", ContextWindow: 202752, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://router.huggingface.co/v1"},
215+ {Provider: "huggingface", ID: "zai-org/GLM-5.1", DisplayName: "zai-org/GLM-5.1", ContextWindow: 202752, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://router.huggingface.co/v1"},
216+ // ----- kimi -----
217+ {Provider: "kimi", ID: "kimi-k2-thinking", DisplayName: "kimi-k2-thinking", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.kimi.com/coding"},
218+ // ----- minimax -----
219+ {Provider: "minimax", ID: "MiniMax-M2.7", DisplayName: "MiniMax-M2.7", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.06, BaseURL: "https://api.minimax.io/anthropic"},
220+ {Provider: "minimax", ID: "MiniMax-M2.7-highspeed", DisplayName: "MiniMax-M2.7-highspeed", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.4, PriceCacheRead: 0.06, BaseURL: "https://api.minimax.io/anthropic"},
221+ // ----- minimax-cn -----
222+ {Provider: "minimax-cn", ID: "MiniMax-M2.7", DisplayName: "MiniMax-M2.7", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.06, BaseURL: "https://api.minimaxi.com/anthropic"},
223+ {Provider: "minimax-cn", ID: "MiniMax-M2.7-highspeed", DisplayName: "MiniMax-M2.7-highspeed", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.4, PriceCacheRead: 0.06, BaseURL: "https://api.minimaxi.com/anthropic"},
224+ // ----- mistral -----
225+ {Provider: "mistral", ID: "codestral-latest", DisplayName: "codestral-latest", ContextWindow: 256000, MaxOutput: 4096, Reasoning: false, PriceInput: 0.3, PriceOutput: 0.9, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
226+ {Provider: "mistral", ID: "devstral-2512", DisplayName: "devstral-2512", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
227+ {Provider: "mistral", ID: "devstral-medium-2507", DisplayName: "devstral-medium-2507", ContextWindow: 128000, MaxOutput: 128000, Reasoning: false, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
228+ {Provider: "mistral", ID: "devstral-medium-latest", DisplayName: "devstral-medium-latest", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
229+ {Provider: "mistral", ID: "devstral-small-2505", DisplayName: "devstral-small-2505", ContextWindow: 128000, MaxOutput: 128000, Reasoning: false, PriceInput: 0.1, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
230+ {Provider: "mistral", ID: "devstral-small-2507", DisplayName: "devstral-small-2507", ContextWindow: 128000, MaxOutput: 128000, Reasoning: false, PriceInput: 0.1, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
231+ {Provider: "mistral", ID: "labs-devstral-small-2512", DisplayName: "labs-devstral-small-2512", ContextWindow: 256000, MaxOutput: 256000, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
232+ {Provider: "mistral", ID: "magistral-medium-latest", DisplayName: "magistral-medium-latest", ContextWindow: 128000, MaxOutput: 16384, Reasoning: true, PriceInput: 2, PriceOutput: 5, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
233+ {Provider: "mistral", ID: "magistral-small", DisplayName: "magistral-small", ContextWindow: 128000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.5, PriceOutput: 1.5, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
234+ {Provider: "mistral", ID: "ministral-3b-latest", DisplayName: "ministral-3b-latest", ContextWindow: 128000, MaxOutput: 128000, Reasoning: false, PriceInput: 0.04, PriceOutput: 0.04, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
235+ {Provider: "mistral", ID: "ministral-8b-latest", DisplayName: "ministral-8b-latest", ContextWindow: 128000, MaxOutput: 128000, Reasoning: false, PriceInput: 0.1, PriceOutput: 0.1, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
236+ {Provider: "mistral", ID: "mistral-large-2411", DisplayName: "mistral-large-2411", ContextWindow: 131072, MaxOutput: 16384, Reasoning: false, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
237+ {Provider: "mistral", ID: "mistral-large-2512", DisplayName: "mistral-large-2512", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 0.5, PriceOutput: 1.5, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
238+ {Provider: "mistral", ID: "mistral-large-latest", DisplayName: "mistral-large-latest", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 0.5, PriceOutput: 1.5, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
239+ {Provider: "mistral", ID: "mistral-medium-2505", DisplayName: "mistral-medium-2505", ContextWindow: 131072, MaxOutput: 131072, Reasoning: false, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
240+ {Provider: "mistral", ID: "mistral-medium-2508", DisplayName: "mistral-medium-2508", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
241+ {Provider: "mistral", ID: "mistral-medium-2604", DisplayName: "mistral-medium-2604", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 1.5, PriceOutput: 7.5, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
242+ {Provider: "mistral", ID: "mistral-medium-3.5", DisplayName: "mistral-medium-3.5", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 1.5, PriceOutput: 7.5, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
243+ {Provider: "mistral", ID: "mistral-medium-latest", DisplayName: "mistral-medium-latest", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 1.5, PriceOutput: 7.5, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
244+ {Provider: "mistral", ID: "mistral-nemo", DisplayName: "mistral-nemo", ContextWindow: 128000, MaxOutput: 128000, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.15, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
245+ {Provider: "mistral", ID: "mistral-small-2506", DisplayName: "mistral-small-2506", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 0.1, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
246+ {Provider: "mistral", ID: "mistral-small-2603", DisplayName: "mistral-small-2603", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
247+ {Provider: "mistral", ID: "mistral-small-latest", DisplayName: "mistral-small-latest", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
248+ {Provider: "mistral", ID: "open-mistral-7b", DisplayName: "open-mistral-7b", ContextWindow: 8000, MaxOutput: 8000, Reasoning: false, PriceInput: 0.25, PriceOutput: 0.25, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
249+ {Provider: "mistral", ID: "open-mixtral-8x22b", DisplayName: "open-mixtral-8x22b", ContextWindow: 64000, MaxOutput: 64000, Reasoning: false, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
250+ {Provider: "mistral", ID: "open-mixtral-8x7b", DisplayName: "open-mixtral-8x7b", ContextWindow: 32000, MaxOutput: 32000, Reasoning: false, PriceInput: 0.7, PriceOutput: 0.7, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
251+ {Provider: "mistral", ID: "pixtral-12b", DisplayName: "pixtral-12b", ContextWindow: 128000, MaxOutput: 128000, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.15, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
252+ {Provider: "mistral", ID: "pixtral-large-latest", DisplayName: "pixtral-large-latest", ContextWindow: 128000, MaxOutput: 128000, Reasoning: false, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0, BaseURL: "https://api.mistral.ai"},
253+ // ----- moonshotai -----
254+ {Provider: "moonshotai", ID: "kimi-k2-0711-preview", DisplayName: "kimi-k2-0711-preview", ContextWindow: 131072, MaxOutput: 16384, Reasoning: false, PriceInput: 0.6, PriceOutput: 2.5, PriceCacheRead: 0.15, BaseURL: "https://api.moonshot.ai/v1"},
255+ {Provider: "moonshotai", ID: "kimi-k2-0905-preview", DisplayName: "kimi-k2-0905-preview", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 0.6, PriceOutput: 2.5, PriceCacheRead: 0.15, BaseURL: "https://api.moonshot.ai/v1"},
256+ {Provider: "moonshotai", ID: "kimi-k2-thinking", DisplayName: "kimi-k2-thinking", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.5, PriceCacheRead: 0.15, BaseURL: "https://api.moonshot.ai/v1"},
257+ {Provider: "moonshotai", ID: "kimi-k2-thinking-turbo", DisplayName: "kimi-k2-thinking-turbo", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 1.15, PriceOutput: 8, PriceCacheRead: 0.15, BaseURL: "https://api.moonshot.ai/v1"},
258+ {Provider: "moonshotai", ID: "kimi-k2-turbo-preview", DisplayName: "kimi-k2-turbo-preview", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 2.4, PriceOutput: 10, PriceCacheRead: 0.6, BaseURL: "https://api.moonshot.ai/v1"},
259+ {Provider: "moonshotai", ID: "kimi-k2.5", DisplayName: "kimi-k2.5", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://api.moonshot.ai/v1"},
260+ {Provider: "moonshotai", ID: "kimi-k2.6", DisplayName: "kimi-k2.6", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://api.moonshot.ai/v1"},
261+ // ----- moonshotai-cn -----
262+ {Provider: "moonshotai-cn", ID: "kimi-k2-0711-preview", DisplayName: "kimi-k2-0711-preview", ContextWindow: 131072, MaxOutput: 16384, Reasoning: false, PriceInput: 0.6, PriceOutput: 2.5, PriceCacheRead: 0.15, BaseURL: "https://api.moonshot.cn/v1"},
263+ {Provider: "moonshotai-cn", ID: "kimi-k2-0905-preview", DisplayName: "kimi-k2-0905-preview", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 0.6, PriceOutput: 2.5, PriceCacheRead: 0.15, BaseURL: "https://api.moonshot.cn/v1"},
264+ {Provider: "moonshotai-cn", ID: "kimi-k2-thinking", DisplayName: "kimi-k2-thinking", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.5, PriceCacheRead: 0.15, BaseURL: "https://api.moonshot.cn/v1"},
265+ {Provider: "moonshotai-cn", ID: "kimi-k2-thinking-turbo", DisplayName: "kimi-k2-thinking-turbo", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 1.15, PriceOutput: 8, PriceCacheRead: 0.15, BaseURL: "https://api.moonshot.cn/v1"},
266+ {Provider: "moonshotai-cn", ID: "kimi-k2-turbo-preview", DisplayName: "kimi-k2-turbo-preview", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 2.4, PriceOutput: 10, PriceCacheRead: 0.6, BaseURL: "https://api.moonshot.cn/v1"},
267+ {Provider: "moonshotai-cn", ID: "kimi-k2.5", DisplayName: "kimi-k2.5", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://api.moonshot.cn/v1"},
268+ {Provider: "moonshotai-cn", ID: "kimi-k2.6", DisplayName: "kimi-k2.6", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://api.moonshot.cn/v1"},
269+ // ----- opencode -----
270+ {Provider: "opencode", ID: "big-pickle", DisplayName: "big-pickle", ContextWindow: 200000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://opencode.ai/zen/v1"},
271+ {Provider: "opencode", ID: "claude-haiku-4-5", DisplayName: "claude-haiku-4-5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.1, BaseURL: "https://opencode.ai/zen"},
272+ {Provider: "opencode", ID: "claude-opus-4-1", DisplayName: "claude-opus-4-1", ContextWindow: 200000, MaxOutput: 32000, Reasoning: true, PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, BaseURL: "https://opencode.ai/zen"},
273+ {Provider: "opencode", ID: "claude-opus-4-5", DisplayName: "claude-opus-4-5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://opencode.ai/zen"},
274+ {Provider: "opencode", ID: "claude-opus-4-6", DisplayName: "claude-opus-4-6", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://opencode.ai/zen"},
275+ {Provider: "opencode", ID: "claude-opus-4-7", DisplayName: "claude-opus-4-7", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://opencode.ai/zen"},
276+ {Provider: "opencode", ID: "claude-sonnet-4", DisplayName: "claude-sonnet-4", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://opencode.ai/zen"},
277+ {Provider: "opencode", ID: "claude-sonnet-4-5", DisplayName: "claude-sonnet-4-5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://opencode.ai/zen"},
278+ {Provider: "opencode", ID: "claude-sonnet-4-6", DisplayName: "claude-sonnet-4-6", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://opencode.ai/zen"},
279+ {Provider: "opencode", ID: "deepseek-v4-flash-free", DisplayName: "deepseek-v4-flash-free", ContextWindow: 200000, MaxOutput: 128000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://opencode.ai/zen/v1"},
280+ {Provider: "opencode", ID: "gemini-3-flash", DisplayName: "gemini-3-flash", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.05, BaseURL: "https://opencode.ai/zen/v1"},
281+ {Provider: "opencode", ID: "gemini-3.1-pro", DisplayName: "gemini-3.1-pro", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/v1"},
282+ {Provider: "opencode", ID: "gemini-3.5-flash", DisplayName: "gemini-3.5-flash", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 1.5, PriceOutput: 9, PriceCacheRead: 0.15, BaseURL: "https://opencode.ai/zen/v1"},
283+ {Provider: "opencode", ID: "glm-5", DisplayName: "glm-5", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/v1"},
284+ {Provider: "opencode", ID: "glm-5.1", DisplayName: "glm-5.1", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://opencode.ai/zen/v1"},
285+ {Provider: "opencode", ID: "gpt-5", DisplayName: "gpt-5", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.07, PriceOutput: 8.5, PriceCacheRead: 0.107, BaseURL: "https://opencode.ai/zen/v1"},
286+ {Provider: "opencode", ID: "gpt-5-codex", DisplayName: "gpt-5-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.07, PriceOutput: 8.5, PriceCacheRead: 0.107, BaseURL: "https://opencode.ai/zen/v1"},
287+ {Provider: "opencode", ID: "gpt-5-nano", DisplayName: "gpt-5-nano", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.05, PriceOutput: 0.4, PriceCacheRead: 0.005, BaseURL: "https://opencode.ai/zen/v1"},
288+ {Provider: "opencode", ID: "gpt-5.1", DisplayName: "gpt-5.1", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.07, PriceOutput: 8.5, PriceCacheRead: 0.107, BaseURL: "https://opencode.ai/zen/v1"},
289+ {Provider: "opencode", ID: "gpt-5.1-codex", DisplayName: "gpt-5.1-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.07, PriceOutput: 8.5, PriceCacheRead: 0.107, BaseURL: "https://opencode.ai/zen/v1"},
290+ // ----- opencode-go -----
291+ {Provider: "opencode-go", ID: "deepseek-v4-flash", DisplayName: "deepseek-v4-flash", ContextWindow: 1000000, MaxOutput: 384000, Reasoning: true, PriceInput: 0.14, PriceOutput: 0.28, PriceCacheRead: 0.0028, BaseURL: "https://opencode.ai/zen/go/v1"},
292+ {Provider: "opencode-go", ID: "deepseek-v4-pro", DisplayName: "deepseek-v4-pro", ContextWindow: 1000000, MaxOutput: 384000, Reasoning: true, PriceInput: 1.74, PriceOutput: 3.48, PriceCacheRead: 0.0145, BaseURL: "https://opencode.ai/zen/go/v1"},
293+ {Provider: "opencode-go", ID: "glm-5", DisplayName: "glm-5", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/go/v1"},
294+ {Provider: "opencode-go", ID: "glm-5.1", DisplayName: "glm-5.1", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://opencode.ai/zen/go/v1"},
295+ {Provider: "opencode-go", ID: "kimi-k2.5", DisplayName: "kimi-k2.5", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://opencode.ai/zen/go/v1"},
296+ {Provider: "opencode-go", ID: "kimi-k2.6", DisplayName: "kimi-k2.6", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://opencode.ai/zen/go/v1"},
297+ {Provider: "opencode-go", ID: "mimo-v2.5", DisplayName: "mimo-v2.5", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://opencode.ai/zen/go/v1"},
298+ {Provider: "opencode-go", ID: "mimo-v2.5-pro", DisplayName: "mimo-v2.5-pro", ContextWindow: 1048576, MaxOutput: 128000, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/go/v1"},
299+ {Provider: "opencode-go", ID: "minimax-m2.5", DisplayName: "minimax-m2.5", ContextWindow: 204800, MaxOutput: 65536, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.03, BaseURL: "https://opencode.ai/zen/go"},
300+ {Provider: "opencode-go", ID: "minimax-m2.7", DisplayName: "minimax-m2.7", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.06, BaseURL: "https://opencode.ai/zen/go/v1"},
301+ {Provider: "opencode-go", ID: "qwen3.5-plus", DisplayName: "qwen3.5-plus", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.2, PriceOutput: 1.2, PriceCacheRead: 0.02, BaseURL: "https://opencode.ai/zen/go/v1"},
302+ {Provider: "opencode-go", ID: "qwen3.6-plus", DisplayName: "qwen3.6-plus", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.05, BaseURL: "https://opencode.ai/zen/go/v1"},
303+ // ----- openrouter -----
304+ {Provider: "openrouter", ID: "ai21/jamba-large-1.7", DisplayName: "ai21/jamba-large-1.7", ContextWindow: 256000, MaxOutput: 4096, Reasoning: false, PriceInput: 2, PriceOutput: 8, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
305+ {Provider: "openrouter", ID: "alibaba/tongyi-deepresearch-30b-a3b", DisplayName: "alibaba/tongyi-deepresearch-30b-a3b", ContextWindow: 131072, MaxOutput: 131072, Reasoning: true, PriceInput: 0.09, PriceOutput: 0.44999999999999996, PriceCacheRead: 0.09, BaseURL: "https://openrouter.ai/api/v1"},
306+ {Provider: "openrouter", ID: "amazon/nova-2-lite-v1", DisplayName: "amazon/nova-2-lite-v1", ContextWindow: 1000000, MaxOutput: 65535, Reasoning: true, PriceInput: 0.3, PriceOutput: 2.5, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
307+ {Provider: "openrouter", ID: "amazon/nova-lite-v1", DisplayName: "amazon/nova-lite-v1", ContextWindow: 300000, MaxOutput: 5120, Reasoning: false, PriceInput: 0.06, PriceOutput: 0.24, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
308+ {Provider: "openrouter", ID: "amazon/nova-micro-v1", DisplayName: "amazon/nova-micro-v1", ContextWindow: 128000, MaxOutput: 5120, Reasoning: false, PriceInput: 0.035, PriceOutput: 0.14, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
309+ {Provider: "openrouter", ID: "amazon/nova-premier-v1", DisplayName: "amazon/nova-premier-v1", ContextWindow: 1000000, MaxOutput: 32000, Reasoning: false, PriceInput: 2.5, PriceOutput: 12.5, PriceCacheRead: 0.625, BaseURL: "https://openrouter.ai/api/v1"},
310+ {Provider: "openrouter", ID: "amazon/nova-pro-v1", DisplayName: "amazon/nova-pro-v1", ContextWindow: 300000, MaxOutput: 5120, Reasoning: false, PriceInput: 0.7999999999999999, PriceOutput: 3.1999999999999997, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
311+ {Provider: "openrouter", ID: "anthropic/claude-3-haiku", DisplayName: "anthropic/claude-3-haiku", ContextWindow: 200000, MaxOutput: 4096, Reasoning: false, PriceInput: 0.25, PriceOutput: 1.25, PriceCacheRead: 0.03, BaseURL: "https://openrouter.ai/api/v1"},
312+ {Provider: "openrouter", ID: "anthropic/claude-3.5-haiku", DisplayName: "anthropic/claude-3.5-haiku", ContextWindow: 200000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.7999999999999999, PriceOutput: 4, PriceCacheRead: 0.08, BaseURL: "https://openrouter.ai/api/v1"},
313+ {Provider: "openrouter", ID: "anthropic/claude-haiku-4.5", DisplayName: "anthropic/claude-haiku-4.5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.09999999999999999, BaseURL: "https://openrouter.ai/api/v1"},
314+ {Provider: "openrouter", ID: "anthropic/claude-opus-4", DisplayName: "anthropic/claude-opus-4", ContextWindow: 200000, MaxOutput: 32000, Reasoning: true, PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, BaseURL: "https://openrouter.ai/api/v1"},
315+ {Provider: "openrouter", ID: "anthropic/claude-opus-4.1", DisplayName: "anthropic/claude-opus-4.1", ContextWindow: 200000, MaxOutput: 32000, Reasoning: true, PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, BaseURL: "https://openrouter.ai/api/v1"},
316+ {Provider: "openrouter", ID: "anthropic/claude-opus-4.5", DisplayName: "anthropic/claude-opus-4.5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://openrouter.ai/api/v1"},
317+ {Provider: "openrouter", ID: "anthropic/claude-opus-4.6", DisplayName: "anthropic/claude-opus-4.6", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://openrouter.ai/api/v1"},
318+ {Provider: "openrouter", ID: "anthropic/claude-opus-4.6-fast", DisplayName: "anthropic/claude-opus-4.6-fast", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 30, PriceOutput: 150, PriceCacheRead: 3, BaseURL: "https://openrouter.ai/api/v1"},
319+ {Provider: "openrouter", ID: "anthropic/claude-opus-4.7", DisplayName: "anthropic/claude-opus-4.7", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://openrouter.ai/api/v1"},
320+ {Provider: "openrouter", ID: "anthropic/claude-opus-4.7-fast", DisplayName: "anthropic/claude-opus-4.7-fast", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 30, PriceOutput: 150, PriceCacheRead: 3, BaseURL: "https://openrouter.ai/api/v1"},
321+ {Provider: "openrouter", ID: "anthropic/claude-sonnet-4", DisplayName: "anthropic/claude-sonnet-4", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://openrouter.ai/api/v1"},
322+ {Provider: "openrouter", ID: "anthropic/claude-sonnet-4.5", DisplayName: "anthropic/claude-sonnet-4.5", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://openrouter.ai/api/v1"},
323+ {Provider: "openrouter", ID: "anthropic/claude-sonnet-4.6", DisplayName: "anthropic/claude-sonnet-4.6", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://openrouter.ai/api/v1"},
324+ {Provider: "openrouter", ID: "arcee-ai/trinity-large-thinking", DisplayName: "arcee-ai/trinity-large-thinking", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.22, PriceOutput: 0.85, PriceCacheRead: 0.06, BaseURL: "https://openrouter.ai/api/v1"},
325+ {Provider: "openrouter", ID: "arcee-ai/trinity-large-thinking:free", DisplayName: "arcee-ai/trinity-large-thinking:free", ContextWindow: 262144, MaxOutput: 80000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
326+ {Provider: "openrouter", ID: "arcee-ai/trinity-mini", DisplayName: "arcee-ai/trinity-mini", ContextWindow: 131072, MaxOutput: 131072, Reasoning: true, PriceInput: 0.045, PriceOutput: 0.15, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
327+ {Provider: "openrouter", ID: "arcee-ai/virtuoso-large", DisplayName: "arcee-ai/virtuoso-large", ContextWindow: 131072, MaxOutput: 64000, Reasoning: false, PriceInput: 0.75, PriceOutput: 1.2, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
328+ {Provider: "openrouter", ID: "auto", DisplayName: "auto", ContextWindow: 2000000, MaxOutput: 30000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
329+ {Provider: "openrouter", ID: "baidu/cobuddy:free", DisplayName: "baidu/cobuddy:free", ContextWindow: 131072, MaxOutput: 65536, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
330+ {Provider: "openrouter", ID: "baidu/ernie-4.5-21b-a3b", DisplayName: "baidu/ernie-4.5-21b-a3b", ContextWindow: 131072, MaxOutput: 8000, Reasoning: false, PriceInput: 0.07, PriceOutput: 0.28, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
331+ {Provider: "openrouter", ID: "baidu/ernie-4.5-vl-28b-a3b", DisplayName: "baidu/ernie-4.5-vl-28b-a3b", ContextWindow: 131072, MaxOutput: 8000, Reasoning: true, PriceInput: 0.14, PriceOutput: 0.56, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
332+ {Provider: "openrouter", ID: "bytedance-seed/seed-1.6", DisplayName: "bytedance-seed/seed-1.6", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0.25, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
333+ {Provider: "openrouter", ID: "bytedance-seed/seed-1.6-flash", DisplayName: "bytedance-seed/seed-1.6-flash", ContextWindow: 262144, MaxOutput: 32768, Reasoning: true, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://openrouter.ai/api/v1"},
334+ // ----- together -----
335+ {Provider: "together", ID: "MiniMaxAI/MiniMax-M2.5", DisplayName: "MiniMaxAI/MiniMax-M2.5", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.06, BaseURL: "https://api.together.ai/v1"},
336+ {Provider: "together", ID: "MiniMaxAI/MiniMax-M2.7", DisplayName: "MiniMaxAI/MiniMax-M2.7", ContextWindow: 202752, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.06, BaseURL: "https://api.together.ai/v1"},
337+ {Provider: "together", ID: "Qwen/Qwen3-235B-A22B-Instruct-2507-tput", DisplayName: "Qwen/Qwen3-235B-A22B-Instruct-2507-tput", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.2, PriceOutput: 0.6, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
338+ {Provider: "together", ID: "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8", DisplayName: "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8", ContextWindow: 262144, MaxOutput: 262144, Reasoning: false, PriceInput: 2, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
339+ {Provider: "together", ID: "Qwen/Qwen3-Coder-Next-FP8", DisplayName: "Qwen/Qwen3-Coder-Next-FP8", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.5, PriceOutput: 1.2, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
340+ {Provider: "together", ID: "Qwen/Qwen3.5-397B-A17B", DisplayName: "Qwen/Qwen3.5-397B-A17B", ContextWindow: 262144, MaxOutput: 130000, Reasoning: true, PriceInput: 0.6, PriceOutput: 3.6, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
341+ {Provider: "together", ID: "Qwen/Qwen3.6-Plus", DisplayName: "Qwen/Qwen3.6-Plus", ContextWindow: 1000000, MaxOutput: 500000, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
342+ {Provider: "together", ID: "Qwen/Qwen3.7-Max", DisplayName: "Qwen/Qwen3.7-Max", ContextWindow: 1000000, MaxOutput: 500000, Reasoning: true, PriceInput: 2.5, PriceOutput: 7.5, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
343+ {Provider: "together", ID: "deepseek-ai/DeepSeek-V3", DisplayName: "deepseek-ai/DeepSeek-V3", ContextWindow: 131072, MaxOutput: 131072, Reasoning: true, PriceInput: 1.25, PriceOutput: 1.25, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
344+ {Provider: "together", ID: "deepseek-ai/DeepSeek-V3-1", DisplayName: "deepseek-ai/DeepSeek-V3-1", ContextWindow: 131072, MaxOutput: 131072, Reasoning: true, PriceInput: 0.6, PriceOutput: 1.7, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
345+ {Provider: "together", ID: "deepseek-ai/DeepSeek-V4-Pro", DisplayName: "deepseek-ai/DeepSeek-V4-Pro", ContextWindow: 512000, MaxOutput: 384000, Reasoning: true, PriceInput: 2.1, PriceOutput: 4.4, PriceCacheRead: 0.2, BaseURL: "https://api.together.ai/v1"},
346+ {Provider: "together", ID: "essentialai/Rnj-1-Instruct", DisplayName: "essentialai/Rnj-1-Instruct", ContextWindow: 32768, MaxOutput: 32768, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.15, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
347+ {Provider: "together", ID: "google/gemma-4-31B-it", DisplayName: "google/gemma-4-31B-it", ContextWindow: 262144, MaxOutput: 131072, Reasoning: true, PriceInput: 0.2, PriceOutput: 0.5, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
348+ {Provider: "together", ID: "meta-llama/Llama-3.3-70B-Instruct-Turbo", DisplayName: "meta-llama/Llama-3.3-70B-Instruct-Turbo", ContextWindow: 131072, MaxOutput: 131072, Reasoning: false, PriceInput: 0.88, PriceOutput: 0.88, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
349+ {Provider: "together", ID: "moonshotai/Kimi-K2.5", DisplayName: "moonshotai/Kimi-K2.5", ContextWindow: 262144, MaxOutput: 262144, Reasoning: true, PriceInput: 0.5, PriceOutput: 2.8, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
350+ {Provider: "together", ID: "moonshotai/Kimi-K2.6", DisplayName: "moonshotai/Kimi-K2.6", ContextWindow: 262144, MaxOutput: 131000, Reasoning: true, PriceInput: 1.2, PriceOutput: 4.5, PriceCacheRead: 0.2, BaseURL: "https://api.together.ai/v1"},
351+ {Provider: "together", ID: "openai/gpt-oss-120b", DisplayName: "openai/gpt-oss-120b", ContextWindow: 131072, MaxOutput: 131072, Reasoning: true, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
352+ {Provider: "together", ID: "zai-org/GLM-5.1", DisplayName: "zai-org/GLM-5.1", ContextWindow: 202752, MaxOutput: 131072, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0, BaseURL: "https://api.together.ai/v1"},
353+ // ----- vercel-ai-gateway -----
354+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen-3-14b", DisplayName: "alibaba/qwen-3-14b", ContextWindow: 40960, MaxOutput: 16384, Reasoning: true, PriceInput: 0.12, PriceOutput: 0.24, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
355+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen-3-235b", DisplayName: "alibaba/qwen-3-235b", ContextWindow: 131000, MaxOutput: 40000, Reasoning: false, PriceInput: 0.6, PriceOutput: 1.2, PriceCacheRead: 0.6, BaseURL: "https://ai-gateway.vercel.sh"},
356+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen-3-30b", DisplayName: "alibaba/qwen-3-30b", ContextWindow: 40960, MaxOutput: 16384, Reasoning: true, PriceInput: 0.08, PriceOutput: 0.29, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
357+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen-3-32b", DisplayName: "alibaba/qwen-3-32b", ContextWindow: 128000, MaxOutput: 8192, Reasoning: true, PriceInput: 0.16, PriceOutput: 0.64, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
358+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen-3.6-max-preview", DisplayName: "alibaba/qwen-3.6-max-preview", ContextWindow: 240000, MaxOutput: 64000, Reasoning: true, PriceInput: 1.3, PriceOutput: 7.8, PriceCacheRead: 0.26, BaseURL: "https://ai-gateway.vercel.sh"},
359+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3-235b-a22b-thinking", DisplayName: "alibaba/qwen3-235b-a22b-thinking", ContextWindow: 131072, MaxOutput: 32768, Reasoning: true, PriceInput: 0.39999999999999997, PriceOutput: 4, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
360+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3-coder", DisplayName: "alibaba/qwen3-coder", ContextWindow: 262144, MaxOutput: 65536, Reasoning: false, PriceInput: 1.5, PriceOutput: 7.5, PriceCacheRead: 0.3, BaseURL: "https://ai-gateway.vercel.sh"},
361+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3-coder-30b-a3b", DisplayName: "alibaba/qwen3-coder-30b-a3b", ContextWindow: 262144, MaxOutput: 8192, Reasoning: true, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
362+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3-coder-next", DisplayName: "alibaba/qwen3-coder-next", ContextWindow: 256000, MaxOutput: 256000, Reasoning: false, PriceInput: 0.5, PriceOutput: 1.2, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
363+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3-coder-plus", DisplayName: "alibaba/qwen3-coder-plus", ContextWindow: 1000000, MaxOutput: 65536, Reasoning: false, PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
364+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3-max", DisplayName: "alibaba/qwen3-max", ContextWindow: 262144, MaxOutput: 32768, Reasoning: false, PriceInput: 1.2, PriceOutput: 6, PriceCacheRead: 0.24, BaseURL: "https://ai-gateway.vercel.sh"},
365+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3-max-preview", DisplayName: "alibaba/qwen3-max-preview", ContextWindow: 262144, MaxOutput: 32768, Reasoning: false, PriceInput: 1.2, PriceOutput: 6, PriceCacheRead: 0.24, BaseURL: "https://ai-gateway.vercel.sh"},
366+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3-max-thinking", DisplayName: "alibaba/qwen3-max-thinking", ContextWindow: 256000, MaxOutput: 65536, Reasoning: true, PriceInput: 1.2, PriceOutput: 6, PriceCacheRead: 0.24, BaseURL: "https://ai-gateway.vercel.sh"},
367+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3-vl-thinking", DisplayName: "alibaba/qwen3-vl-thinking", ContextWindow: 131072, MaxOutput: 32768, Reasoning: true, PriceInput: 0.39999999999999997, PriceOutput: 4, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
368+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3.5-flash", DisplayName: "alibaba/qwen3.5-flash", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0.09999999999999999, PriceOutput: 0.39999999999999997, PriceCacheRead: 0.001, BaseURL: "https://ai-gateway.vercel.sh"},
369+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3.5-plus", DisplayName: "alibaba/qwen3.5-plus", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0.39999999999999997, PriceOutput: 2.4, PriceCacheRead: 0.04, BaseURL: "https://ai-gateway.vercel.sh"},
370+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3.6-27b", DisplayName: "alibaba/qwen3.6-27b", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 0.6, PriceOutput: 3.5999999999999996, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
371+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3.6-plus", DisplayName: "alibaba/qwen3.6-plus", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.09999999999999999, BaseURL: "https://ai-gateway.vercel.sh"},
372+ {Provider: "vercel-ai-gateway", ID: "alibaba/qwen3.7-max", DisplayName: "alibaba/qwen3.7-max", ContextWindow: 991000, MaxOutput: 64000, Reasoning: true, PriceInput: 1.25, PriceOutput: 3.75, PriceCacheRead: 0.25, BaseURL: "https://ai-gateway.vercel.sh"},
373+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-3-haiku", DisplayName: "anthropic/claude-3-haiku", ContextWindow: 200000, MaxOutput: 4096, Reasoning: false, PriceInput: 0.25, PriceOutput: 1.25, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
374+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-3.5-haiku", DisplayName: "anthropic/claude-3.5-haiku", ContextWindow: 200000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.7999999999999999, PriceOutput: 4, PriceCacheRead: 0.08, BaseURL: "https://ai-gateway.vercel.sh"},
375+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-haiku-4.5", DisplayName: "anthropic/claude-haiku-4.5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.09999999999999999, BaseURL: "https://ai-gateway.vercel.sh"},
376+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-opus-4", DisplayName: "anthropic/claude-opus-4", ContextWindow: 200000, MaxOutput: 32000, Reasoning: true, PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, BaseURL: "https://ai-gateway.vercel.sh"},
377+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-opus-4.1", DisplayName: "anthropic/claude-opus-4.1", ContextWindow: 200000, MaxOutput: 32000, Reasoning: true, PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, BaseURL: "https://ai-gateway.vercel.sh"},
378+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-opus-4.5", DisplayName: "anthropic/claude-opus-4.5", ContextWindow: 200000, MaxOutput: 64000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://ai-gateway.vercel.sh"},
379+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-opus-4.6", DisplayName: "anthropic/claude-opus-4.6", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://ai-gateway.vercel.sh"},
380+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-opus-4.7", DisplayName: "anthropic/claude-opus-4.7", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, BaseURL: "https://ai-gateway.vercel.sh"},
381+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-sonnet-4", DisplayName: "anthropic/claude-sonnet-4", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://ai-gateway.vercel.sh"},
382+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-sonnet-4.5", DisplayName: "anthropic/claude-sonnet-4.5", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://ai-gateway.vercel.sh"},
383+ {Provider: "vercel-ai-gateway", ID: "anthropic/claude-sonnet-4.6", DisplayName: "anthropic/claude-sonnet-4.6", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, BaseURL: "https://ai-gateway.vercel.sh"},
384+ {Provider: "vercel-ai-gateway", ID: "arcee-ai/trinity-large-preview", DisplayName: "arcee-ai/trinity-large-preview", ContextWindow: 131000, MaxOutput: 131000, Reasoning: false, PriceInput: 0.25, PriceOutput: 1, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
385+ {Provider: "vercel-ai-gateway", ID: "arcee-ai/trinity-large-thinking", DisplayName: "arcee-ai/trinity-large-thinking", ContextWindow: 262100, MaxOutput: 80000, Reasoning: true, PriceInput: 0.25, PriceOutput: 0.8999999999999999, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
386+ {Provider: "vercel-ai-gateway", ID: "bytedance/seed-1.6", DisplayName: "bytedance/seed-1.6", ContextWindow: 256000, MaxOutput: 32000, Reasoning: true, PriceInput: 0.25, PriceOutput: 2, PriceCacheRead: 0.049999999999999996, BaseURL: "https://ai-gateway.vercel.sh"},
387+ {Provider: "vercel-ai-gateway", ID: "cohere/command-a", DisplayName: "cohere/command-a", ContextWindow: 256000, MaxOutput: 8000, Reasoning: false, PriceInput: 2.5, PriceOutput: 10, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
388+ {Provider: "vercel-ai-gateway", ID: "deepseek/deepseek-r1", DisplayName: "deepseek/deepseek-r1", ContextWindow: 128000, MaxOutput: 8192, Reasoning: true, PriceInput: 1.35, PriceOutput: 5.4, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
389+ {Provider: "vercel-ai-gateway", ID: "deepseek/deepseek-v3", DisplayName: "deepseek/deepseek-v3", ContextWindow: 163840, MaxOutput: 16384, Reasoning: false, PriceInput: 0.77, PriceOutput: 0.77, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
390+ {Provider: "vercel-ai-gateway", ID: "deepseek/deepseek-v3.1", DisplayName: "deepseek/deepseek-v3.1", ContextWindow: 163840, MaxOutput: 8192, Reasoning: true, PriceInput: 0.56, PriceOutput: 1.68, PriceCacheRead: 0.28, BaseURL: "https://ai-gateway.vercel.sh"},
391+ {Provider: "vercel-ai-gateway", ID: "deepseek/deepseek-v3.1-terminus", DisplayName: "deepseek/deepseek-v3.1-terminus", ContextWindow: 131072, MaxOutput: 65536, Reasoning: true, PriceInput: 0.27, PriceOutput: 1, PriceCacheRead: 0.135, BaseURL: "https://ai-gateway.vercel.sh"},
392+ {Provider: "vercel-ai-gateway", ID: "deepseek/deepseek-v3.2", DisplayName: "deepseek/deepseek-v3.2", ContextWindow: 128000, MaxOutput: 8000, Reasoning: false, PriceInput: 0.28, PriceOutput: 0.42, PriceCacheRead: 0.028, BaseURL: "https://ai-gateway.vercel.sh"},
393+ {Provider: "vercel-ai-gateway", ID: "deepseek/deepseek-v3.2-thinking", DisplayName: "deepseek/deepseek-v3.2-thinking", ContextWindow: 128000, MaxOutput: 8000, Reasoning: false, PriceInput: 0.62, PriceOutput: 1.85, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
394+ {Provider: "vercel-ai-gateway", ID: "deepseek/deepseek-v4-flash", DisplayName: "deepseek/deepseek-v4-flash", ContextWindow: 1000000, MaxOutput: 384000, Reasoning: true, PriceInput: 0.14, PriceOutput: 0.28, PriceCacheRead: 0.0028, BaseURL: "https://ai-gateway.vercel.sh"},
395+ {Provider: "vercel-ai-gateway", ID: "deepseek/deepseek-v4-pro", DisplayName: "deepseek/deepseek-v4-pro", ContextWindow: 1000000, MaxOutput: 384000, Reasoning: true, PriceInput: 0.435, PriceOutput: 0.87, PriceCacheRead: 0.0036, BaseURL: "https://ai-gateway.vercel.sh"},
396+ {Provider: "vercel-ai-gateway", ID: "google/gemini-2.0-flash", DisplayName: "google/gemini-2.0-flash", ContextWindow: 1048576, MaxOutput: 8192, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0.024999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
397+ {Provider: "vercel-ai-gateway", ID: "google/gemini-2.0-flash-lite", DisplayName: "google/gemini-2.0-flash-lite", ContextWindow: 1048576, MaxOutput: 8192, Reasoning: false, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0.02, BaseURL: "https://ai-gateway.vercel.sh"},
398+ {Provider: "vercel-ai-gateway", ID: "google/gemini-2.5-flash", DisplayName: "google/gemini-2.5-flash", ContextWindow: 1000000, MaxOutput: 65536, Reasoning: true, PriceInput: 0.3, PriceOutput: 2.5, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
399+ {Provider: "vercel-ai-gateway", ID: "google/gemini-2.5-flash-lite", DisplayName: "google/gemini-2.5-flash-lite", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 0.09999999999999999, PriceOutput: 0.39999999999999997, PriceCacheRead: 0.01, BaseURL: "https://ai-gateway.vercel.sh"},
400+ {Provider: "vercel-ai-gateway", ID: "google/gemini-2.5-pro", DisplayName: "google/gemini-2.5-pro", ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://ai-gateway.vercel.sh"},
401+ {Provider: "vercel-ai-gateway", ID: "google/gemini-3-flash", DisplayName: "google/gemini-3-flash", ContextWindow: 1000000, MaxOutput: 65000, Reasoning: true, PriceInput: 0.5, PriceOutput: 3, PriceCacheRead: 0.049999999999999996, BaseURL: "https://ai-gateway.vercel.sh"},
402+ {Provider: "vercel-ai-gateway", ID: "google/gemini-3-pro-preview", DisplayName: "google/gemini-3-pro-preview", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
403+ {Provider: "vercel-ai-gateway", ID: "google/gemini-3.1-flash-lite", DisplayName: "google/gemini-3.1-flash-lite", ContextWindow: 1000000, MaxOutput: 65000, Reasoning: true, PriceInput: 0.25, PriceOutput: 1.5, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
404+ {Provider: "vercel-ai-gateway", ID: "google/gemini-3.1-flash-lite-preview", DisplayName: "google/gemini-3.1-flash-lite-preview", ContextWindow: 1000000, MaxOutput: 65000, Reasoning: true, PriceInput: 0.25, PriceOutput: 1.5, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
405+ {Provider: "vercel-ai-gateway", ID: "google/gemini-3.1-pro-preview", DisplayName: "google/gemini-3.1-pro-preview", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 2, PriceOutput: 12, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
406+ {Provider: "vercel-ai-gateway", ID: "google/gemini-3.5-flash", DisplayName: "google/gemini-3.5-flash", ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true, PriceInput: 1.5, PriceOutput: 9, PriceCacheRead: 0.15, BaseURL: "https://ai-gateway.vercel.sh"},
407+ {Provider: "vercel-ai-gateway", ID: "google/gemma-4-26b-a4b-it", DisplayName: "google/gemma-4-26b-a4b-it", ContextWindow: 262144, MaxOutput: 131072, Reasoning: false, PriceInput: 0.13, PriceOutput: 0.39999999999999997, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
408+ {Provider: "vercel-ai-gateway", ID: "google/gemma-4-31b-it", DisplayName: "google/gemma-4-31b-it", ContextWindow: 262144, MaxOutput: 131072, Reasoning: false, PriceInput: 0.14, PriceOutput: 0.39999999999999997, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
409+ {Provider: "vercel-ai-gateway", ID: "inception/mercury-2", DisplayName: "inception/mercury-2", ContextWindow: 128000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.25, PriceOutput: 0.75, PriceCacheRead: 0.024999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
410+ {Provider: "vercel-ai-gateway", ID: "inception/mercury-coder-small", DisplayName: "inception/mercury-coder-small", ContextWindow: 32000, MaxOutput: 16384, Reasoning: false, PriceInput: 0.25, PriceOutput: 1, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
411+ {Provider: "vercel-ai-gateway", ID: "kwaipilot/kat-coder-pro-v2", DisplayName: "kwaipilot/kat-coder-pro-v2", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.06, BaseURL: "https://ai-gateway.vercel.sh"},
412+ {Provider: "vercel-ai-gateway", ID: "meituan/longcat-flash-chat", DisplayName: "meituan/longcat-flash-chat", ContextWindow: 128000, MaxOutput: 100000, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
413+ {Provider: "vercel-ai-gateway", ID: "meta/llama-3.1-70b", DisplayName: "meta/llama-3.1-70b", ContextWindow: 128000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.72, PriceOutput: 0.72, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
414+ {Provider: "vercel-ai-gateway", ID: "meta/llama-3.1-8b", DisplayName: "meta/llama-3.1-8b", ContextWindow: 128000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.22, PriceOutput: 0.22, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
415+ {Provider: "vercel-ai-gateway", ID: "meta/llama-3.2-11b", DisplayName: "meta/llama-3.2-11b", ContextWindow: 128000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.16, PriceOutput: 0.16, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
416+ {Provider: "vercel-ai-gateway", ID: "meta/llama-3.2-90b", DisplayName: "meta/llama-3.2-90b", ContextWindow: 128000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.72, PriceOutput: 0.72, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
417+ {Provider: "vercel-ai-gateway", ID: "meta/llama-3.3-70b", DisplayName: "meta/llama-3.3-70b", ContextWindow: 128000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.72, PriceOutput: 0.72, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
418+ {Provider: "vercel-ai-gateway", ID: "meta/llama-4-maverick", DisplayName: "meta/llama-4-maverick", ContextWindow: 128000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.24, PriceOutput: 0.9700000000000001, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
419+ {Provider: "vercel-ai-gateway", ID: "meta/llama-4-scout", DisplayName: "meta/llama-4-scout", ContextWindow: 128000, MaxOutput: 8192, Reasoning: false, PriceInput: 0.16999999999999998, PriceOutput: 0.66, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
420+ {Provider: "vercel-ai-gateway", ID: "minimax/minimax-m2", DisplayName: "minimax/minimax-m2", ContextWindow: 205000, MaxOutput: 205000, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
421+ {Provider: "vercel-ai-gateway", ID: "minimax/minimax-m2.1", DisplayName: "minimax/minimax-m2.1", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
422+ {Provider: "vercel-ai-gateway", ID: "minimax/minimax-m2.1-lightning", DisplayName: "minimax/minimax-m2.1-lightning", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0.3, PriceOutput: 2.4, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
423+ {Provider: "vercel-ai-gateway", ID: "minimax/minimax-m2.5", DisplayName: "minimax/minimax-m2.5", ContextWindow: 204800, MaxOutput: 131000, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
424+ {Provider: "vercel-ai-gateway", ID: "minimax/minimax-m2.5-highspeed", DisplayName: "minimax/minimax-m2.5-highspeed", ContextWindow: 204800, MaxOutput: 131000, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.4, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
425+ {Provider: "vercel-ai-gateway", ID: "minimax/minimax-m2.7", DisplayName: "minimax/minimax-m2.7", ContextWindow: 204800, MaxOutput: 131000, Reasoning: true, PriceInput: 0.3, PriceOutput: 1.2, PriceCacheRead: 0.06, BaseURL: "https://ai-gateway.vercel.sh"},
426+ {Provider: "vercel-ai-gateway", ID: "minimax/minimax-m2.7-highspeed", DisplayName: "minimax/minimax-m2.7-highspeed", ContextWindow: 204800, MaxOutput: 131100, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.4, PriceCacheRead: 0.06, BaseURL: "https://ai-gateway.vercel.sh"},
427+ {Provider: "vercel-ai-gateway", ID: "mistral/codestral", DisplayName: "mistral/codestral", ContextWindow: 128000, MaxOutput: 4000, Reasoning: false, PriceInput: 0.3, PriceOutput: 0.8999999999999999, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
428+ {Provider: "vercel-ai-gateway", ID: "mistral/devstral-2", DisplayName: "mistral/devstral-2", ContextWindow: 256000, MaxOutput: 256000, Reasoning: false, PriceInput: 0.39999999999999997, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
429+ {Provider: "vercel-ai-gateway", ID: "mistral/devstral-small", DisplayName: "mistral/devstral-small", ContextWindow: 128000, MaxOutput: 64000, Reasoning: false, PriceInput: 0.09999999999999999, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
430+ {Provider: "vercel-ai-gateway", ID: "mistral/devstral-small-2", DisplayName: "mistral/devstral-small-2", ContextWindow: 256000, MaxOutput: 256000, Reasoning: false, PriceInput: 0.09999999999999999, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
431+ {Provider: "vercel-ai-gateway", ID: "mistral/ministral-3b", DisplayName: "mistral/ministral-3b", ContextWindow: 128000, MaxOutput: 4000, Reasoning: false, PriceInput: 0.09999999999999999, PriceOutput: 0.09999999999999999, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
432+ {Provider: "vercel-ai-gateway", ID: "mistral/ministral-8b", DisplayName: "mistral/ministral-8b", ContextWindow: 128000, MaxOutput: 4000, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.15, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
433+ {Provider: "vercel-ai-gateway", ID: "mistral/mistral-medium", DisplayName: "mistral/mistral-medium", ContextWindow: 128000, MaxOutput: 64000, Reasoning: false, PriceInput: 0.39999999999999997, PriceOutput: 2, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
434+ {Provider: "vercel-ai-gateway", ID: "mistral/mistral-medium-3.5", DisplayName: "mistral/mistral-medium-3.5", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 1.5, PriceOutput: 7.5, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
435+ {Provider: "vercel-ai-gateway", ID: "mistral/mistral-small", DisplayName: "mistral/mistral-small", ContextWindow: 32000, MaxOutput: 4000, Reasoning: false, PriceInput: 0.09999999999999999, PriceOutput: 0.3, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
436+ {Provider: "vercel-ai-gateway", ID: "mistral/pixtral-12b", DisplayName: "mistral/pixtral-12b", ContextWindow: 128000, MaxOutput: 4000, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.15, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
437+ {Provider: "vercel-ai-gateway", ID: "mistral/pixtral-large", DisplayName: "mistral/pixtral-large", ContextWindow: 128000, MaxOutput: 4000, Reasoning: false, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
438+ {Provider: "vercel-ai-gateway", ID: "moonshotai/kimi-k2", DisplayName: "moonshotai/kimi-k2", ContextWindow: 131072, MaxOutput: 131072, Reasoning: false, PriceInput: 0.5700000000000001, PriceOutput: 2.3, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
439+ {Provider: "vercel-ai-gateway", ID: "moonshotai/kimi-k2-thinking", DisplayName: "moonshotai/kimi-k2-thinking", ContextWindow: 262114, MaxOutput: 262114, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.5, PriceCacheRead: 0.15, BaseURL: "https://ai-gateway.vercel.sh"},
440+ {Provider: "vercel-ai-gateway", ID: "moonshotai/kimi-k2-thinking-turbo", DisplayName: "moonshotai/kimi-k2-thinking-turbo", ContextWindow: 262114, MaxOutput: 262114, Reasoning: true, PriceInput: 1.15, PriceOutput: 8, PriceCacheRead: 0.15, BaseURL: "https://ai-gateway.vercel.sh"},
441+ {Provider: "vercel-ai-gateway", ID: "moonshotai/kimi-k2-turbo", DisplayName: "moonshotai/kimi-k2-turbo", ContextWindow: 256000, MaxOutput: 16384, Reasoning: false, PriceInput: 1.15, PriceOutput: 8, PriceCacheRead: 0.15, BaseURL: "https://ai-gateway.vercel.sh"},
442+ {Provider: "vercel-ai-gateway", ID: "moonshotai/kimi-k2.5", DisplayName: "moonshotai/kimi-k2.5", ContextWindow: 262114, MaxOutput: 262114, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.09999999999999999, BaseURL: "https://ai-gateway.vercel.sh"},
443+ {Provider: "vercel-ai-gateway", ID: "moonshotai/kimi-k2.6", DisplayName: "moonshotai/kimi-k2.6", ContextWindow: 262000, MaxOutput: 262000, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://ai-gateway.vercel.sh"},
444+ {Provider: "vercel-ai-gateway", ID: "nvidia/nemotron-nano-12b-v2-vl", DisplayName: "nvidia/nemotron-nano-12b-v2-vl", ContextWindow: 131072, MaxOutput: 131072, Reasoning: true, PriceInput: 0.19999999999999998, PriceOutput: 0.6, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
445+ {Provider: "vercel-ai-gateway", ID: "nvidia/nemotron-nano-9b-v2", DisplayName: "nvidia/nemotron-nano-9b-v2", ContextWindow: 131072, MaxOutput: 131072, Reasoning: true, PriceInput: 0.06, PriceOutput: 0.22999999999999998, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
446+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-4-turbo", DisplayName: "openai/gpt-4-turbo", ContextWindow: 128000, MaxOutput: 4096, Reasoning: false, PriceInput: 10, PriceOutput: 30, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
447+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-4.1", DisplayName: "openai/gpt-4.1", ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false, PriceInput: 2, PriceOutput: 8, PriceCacheRead: 0.5, BaseURL: "https://ai-gateway.vercel.sh"},
448+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-4.1-mini", DisplayName: "openai/gpt-4.1-mini", ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false, PriceInput: 0.39999999999999997, PriceOutput: 1.5999999999999999, PriceCacheRead: 0.09999999999999999, BaseURL: "https://ai-gateway.vercel.sh"},
449+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-4.1-nano", DisplayName: "openai/gpt-4.1-nano", ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false, PriceInput: 0.09999999999999999, PriceOutput: 0.39999999999999997, PriceCacheRead: 0.024999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
450+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-4o", DisplayName: "openai/gpt-4o", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 2.5, PriceOutput: 10, PriceCacheRead: 1.25, BaseURL: "https://ai-gateway.vercel.sh"},
451+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-4o-mini", DisplayName: "openai/gpt-4o-mini", ContextWindow: 128000, MaxOutput: 16384, Reasoning: false, PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0.075, BaseURL: "https://ai-gateway.vercel.sh"},
452+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5", DisplayName: "openai/gpt-5", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://ai-gateway.vercel.sh"},
453+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5-chat", DisplayName: "openai/gpt-5-chat", ContextWindow: 128000, MaxOutput: 16384, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://ai-gateway.vercel.sh"},
454+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5-codex", DisplayName: "openai/gpt-5-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://ai-gateway.vercel.sh"},
455+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5-mini", DisplayName: "openai/gpt-5-mini", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.25, PriceOutput: 2, PriceCacheRead: 0.024999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
456+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5-nano", DisplayName: "openai/gpt-5-nano", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.049999999999999996, PriceOutput: 0.39999999999999997, PriceCacheRead: 0.005, BaseURL: "https://ai-gateway.vercel.sh"},
457+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5-pro", DisplayName: "openai/gpt-5-pro", ContextWindow: 400000, MaxOutput: 272000, Reasoning: true, PriceInput: 15, PriceOutput: 120, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
458+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.1-codex", DisplayName: "openai/gpt-5.1-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://ai-gateway.vercel.sh"},
459+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.1-codex-max", DisplayName: "openai/gpt-5.1-codex-max", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://ai-gateway.vercel.sh"},
460+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.1-codex-mini", DisplayName: "openai/gpt-5.1-codex-mini", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.25, PriceOutput: 2, PriceCacheRead: 0.024999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
461+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.1-instant", DisplayName: "openai/gpt-5.1-instant", ContextWindow: 128000, MaxOutput: 16384, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://ai-gateway.vercel.sh"},
462+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.1-thinking", DisplayName: "openai/gpt-5.1-thinking", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125, BaseURL: "https://ai-gateway.vercel.sh"},
463+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.2", DisplayName: "openai/gpt-5.2", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175, BaseURL: "https://ai-gateway.vercel.sh"},
464+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.2-chat", DisplayName: "openai/gpt-5.2-chat", ContextWindow: 128000, MaxOutput: 16384, Reasoning: true, PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175, BaseURL: "https://ai-gateway.vercel.sh"},
465+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.2-codex", DisplayName: "openai/gpt-5.2-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175, BaseURL: "https://ai-gateway.vercel.sh"},
466+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.2-pro", DisplayName: "openai/gpt-5.2-pro", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 21, PriceOutput: 168, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
467+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.3-chat", DisplayName: "openai/gpt-5.3-chat", ContextWindow: 128000, MaxOutput: 16384, Reasoning: true, PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175, BaseURL: "https://ai-gateway.vercel.sh"},
468+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.3-codex", DisplayName: "openai/gpt-5.3-codex", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175, BaseURL: "https://ai-gateway.vercel.sh"},
469+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.4", DisplayName: "openai/gpt-5.4", ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true, PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25, BaseURL: "https://ai-gateway.vercel.sh"},
470+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.4-mini", DisplayName: "openai/gpt-5.4-mini", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075, BaseURL: "https://ai-gateway.vercel.sh"},
471+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.4-nano", DisplayName: "openai/gpt-5.4-nano", ContextWindow: 400000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.19999999999999998, PriceOutput: 1.25, PriceCacheRead: 0.02, BaseURL: "https://ai-gateway.vercel.sh"},
472+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.4-pro", DisplayName: "openai/gpt-5.4-pro", ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true, PriceInput: 30, PriceOutput: 180, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
473+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.5", DisplayName: "openai/gpt-5.5", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5, BaseURL: "https://ai-gateway.vercel.sh"},
474+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-5.5-pro", DisplayName: "openai/gpt-5.5-pro", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 30, PriceOutput: 180, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
475+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-oss-20b", DisplayName: "openai/gpt-oss-20b", ContextWindow: 131072, MaxOutput: 8192, Reasoning: true, PriceInput: 0.049999999999999996, PriceOutput: 0.19999999999999998, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
476+ {Provider: "vercel-ai-gateway", ID: "openai/gpt-oss-safeguard-20b", DisplayName: "openai/gpt-oss-safeguard-20b", ContextWindow: 131072, MaxOutput: 65536, Reasoning: true, PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0.037, BaseURL: "https://ai-gateway.vercel.sh"},
477+ {Provider: "vercel-ai-gateway", ID: "openai/o1", DisplayName: "openai/o1", ContextWindow: 200000, MaxOutput: 100000, Reasoning: true, PriceInput: 15, PriceOutput: 60, PriceCacheRead: 7.5, BaseURL: "https://ai-gateway.vercel.sh"},
478+ {Provider: "vercel-ai-gateway", ID: "openai/o3", DisplayName: "openai/o3", ContextWindow: 200000, MaxOutput: 100000, Reasoning: true, PriceInput: 2, PriceOutput: 8, PriceCacheRead: 0.5, BaseURL: "https://ai-gateway.vercel.sh"},
479+ {Provider: "vercel-ai-gateway", ID: "openai/o3-deep-research", DisplayName: "openai/o3-deep-research", ContextWindow: 200000, MaxOutput: 100000, Reasoning: true, PriceInput: 10, PriceOutput: 40, PriceCacheRead: 2.5, BaseURL: "https://ai-gateway.vercel.sh"},
480+ {Provider: "vercel-ai-gateway", ID: "openai/o3-mini", DisplayName: "openai/o3-mini", ContextWindow: 200000, MaxOutput: 100000, Reasoning: true, PriceInput: 1.1, PriceOutput: 4.4, PriceCacheRead: 0.55, BaseURL: "https://ai-gateway.vercel.sh"},
481+ {Provider: "vercel-ai-gateway", ID: "openai/o3-pro", DisplayName: "openai/o3-pro", ContextWindow: 200000, MaxOutput: 100000, Reasoning: true, PriceInput: 20, PriceOutput: 80, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
482+ {Provider: "vercel-ai-gateway", ID: "openai/o4-mini", DisplayName: "openai/o4-mini", ContextWindow: 200000, MaxOutput: 100000, Reasoning: true, PriceInput: 1.1, PriceOutput: 4.4, PriceCacheRead: 0.275, BaseURL: "https://ai-gateway.vercel.sh"},
483+ {Provider: "vercel-ai-gateway", ID: "perplexity/sonar", DisplayName: "perplexity/sonar", ContextWindow: 127000, MaxOutput: 8000, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
484+ {Provider: "vercel-ai-gateway", ID: "perplexity/sonar-pro", DisplayName: "perplexity/sonar-pro", ContextWindow: 200000, MaxOutput: 8000, Reasoning: false, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
485+ {Provider: "vercel-ai-gateway", ID: "xai/grok-4.1-fast-non-reasoning", DisplayName: "xai/grok-4.1-fast-non-reasoning", ContextWindow: 1000000, MaxOutput: 1000000, Reasoning: false, PriceInput: 0.19999999999999998, PriceOutput: 0.5, PriceCacheRead: 0.049999999999999996, BaseURL: "https://ai-gateway.vercel.sh"},
486+ {Provider: "vercel-ai-gateway", ID: "xai/grok-4.1-fast-reasoning", DisplayName: "xai/grok-4.1-fast-reasoning", ContextWindow: 1000000, MaxOutput: 1000000, Reasoning: true, PriceInput: 0.19999999999999998, PriceOutput: 0.5, PriceCacheRead: 0.049999999999999996, BaseURL: "https://ai-gateway.vercel.sh"},
487+ {Provider: "vercel-ai-gateway", ID: "xai/grok-4.20-multi-agent", DisplayName: "xai/grok-4.20-multi-agent", ContextWindow: 2000000, MaxOutput: 2000000, Reasoning: true, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
488+ {Provider: "vercel-ai-gateway", ID: "xai/grok-4.20-multi-agent-beta", DisplayName: "xai/grok-4.20-multi-agent-beta", ContextWindow: 2000000, MaxOutput: 2000000, Reasoning: true, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
489+ {Provider: "vercel-ai-gateway", ID: "xai/grok-4.20-non-reasoning", DisplayName: "xai/grok-4.20-non-reasoning", ContextWindow: 2000000, MaxOutput: 2000000, Reasoning: false, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
490+ {Provider: "vercel-ai-gateway", ID: "xai/grok-4.20-non-reasoning-beta", DisplayName: "xai/grok-4.20-non-reasoning-beta", ContextWindow: 2000000, MaxOutput: 2000000, Reasoning: false, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
491+ {Provider: "vercel-ai-gateway", ID: "xai/grok-4.20-reasoning", DisplayName: "xai/grok-4.20-reasoning", ContextWindow: 2000000, MaxOutput: 2000000, Reasoning: true, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
492+ {Provider: "vercel-ai-gateway", ID: "xai/grok-4.20-reasoning-beta", DisplayName: "xai/grok-4.20-reasoning-beta", ContextWindow: 2000000, MaxOutput: 2000000, Reasoning: true, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
493+ {Provider: "vercel-ai-gateway", ID: "xai/grok-4.3", DisplayName: "xai/grok-4.3", ContextWindow: 1000000, MaxOutput: 1000000, Reasoning: true, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
494+ {Provider: "vercel-ai-gateway", ID: "xai/grok-build-0.1", DisplayName: "xai/grok-build-0.1", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 1, PriceOutput: 2, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
495+ {Provider: "vercel-ai-gateway", ID: "xiaomi/mimo-v2-flash", DisplayName: "xiaomi/mimo-v2-flash", ContextWindow: 262144, MaxOutput: 32000, Reasoning: true, PriceInput: 0.09999999999999999, PriceOutput: 0.3, PriceCacheRead: 0.01, BaseURL: "https://ai-gateway.vercel.sh"},
496+ {Provider: "vercel-ai-gateway", ID: "xiaomi/mimo-v2-pro", DisplayName: "xiaomi/mimo-v2-pro", ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
497+ {Provider: "vercel-ai-gateway", ID: "xiaomi/mimo-v2.5", DisplayName: "xiaomi/mimo-v2.5", ContextWindow: 1050000, MaxOutput: 131100, Reasoning: true, PriceInput: 0.39999999999999997, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://ai-gateway.vercel.sh"},
498+ {Provider: "vercel-ai-gateway", ID: "xiaomi/mimo-v2.5-pro", DisplayName: "xiaomi/mimo-v2.5-pro", ContextWindow: 1050000, MaxOutput: 131000, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
499+ {Provider: "vercel-ai-gateway", ID: "zai/glm-4.5", DisplayName: "zai/glm-4.5", ContextWindow: 128000, MaxOutput: 96000, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.2, PriceCacheRead: 0.11, BaseURL: "https://ai-gateway.vercel.sh"},
500+ {Provider: "vercel-ai-gateway", ID: "zai/glm-4.5-air", DisplayName: "zai/glm-4.5-air", ContextWindow: 128000, MaxOutput: 96000, Reasoning: true, PriceInput: 0.19999999999999998, PriceOutput: 1.1, PriceCacheRead: 0.03, BaseURL: "https://ai-gateway.vercel.sh"},
501+ {Provider: "vercel-ai-gateway", ID: "zai/glm-4.5v", DisplayName: "zai/glm-4.5v", ContextWindow: 66000, MaxOutput: 16000, Reasoning: false, PriceInput: 0.6, PriceOutput: 1.7999999999999998, PriceCacheRead: 0.11, BaseURL: "https://ai-gateway.vercel.sh"},
502+ {Provider: "vercel-ai-gateway", ID: "zai/glm-4.6", DisplayName: "zai/glm-4.6", ContextWindow: 200000, MaxOutput: 96000, Reasoning: true, PriceInput: 0.6, PriceOutput: 2.2, PriceCacheRead: 0.11, BaseURL: "https://ai-gateway.vercel.sh"},
503+ {Provider: "vercel-ai-gateway", ID: "zai/glm-4.6v", DisplayName: "zai/glm-4.6v", ContextWindow: 128000, MaxOutput: 24000, Reasoning: true, PriceInput: 0.3, PriceOutput: 0.8999999999999999, PriceCacheRead: 0.049999999999999996, BaseURL: "https://ai-gateway.vercel.sh"},
504+ {Provider: "vercel-ai-gateway", ID: "zai/glm-4.6v-flash", DisplayName: "zai/glm-4.6v-flash", ContextWindow: 128000, MaxOutput: 24000, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
505+ {Provider: "vercel-ai-gateway", ID: "zai/glm-4.7", DisplayName: "zai/glm-4.7", ContextWindow: 131000, MaxOutput: 40000, Reasoning: true, PriceInput: 2.25, PriceOutput: 2.75, PriceCacheRead: 2.25, BaseURL: "https://ai-gateway.vercel.sh"},
506+ {Provider: "vercel-ai-gateway", ID: "zai/glm-4.7-flash", DisplayName: "zai/glm-4.7-flash", ContextWindow: 200000, MaxOutput: 131000, Reasoning: true, PriceInput: 0.07, PriceOutput: 0.39999999999999997, PriceCacheRead: 0, BaseURL: "https://ai-gateway.vercel.sh"},
507+ {Provider: "vercel-ai-gateway", ID: "zai/glm-4.7-flashx", DisplayName: "zai/glm-4.7-flashx", ContextWindow: 200000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.06, PriceOutput: 0.39999999999999997, PriceCacheRead: 0.01, BaseURL: "https://ai-gateway.vercel.sh"},
508+ {Provider: "vercel-ai-gateway", ID: "zai/glm-5", DisplayName: "zai/glm-5", ContextWindow: 202800, MaxOutput: 131100, Reasoning: true, PriceInput: 1, PriceOutput: 3.1999999999999997, PriceCacheRead: 0.19999999999999998, BaseURL: "https://ai-gateway.vercel.sh"},
509+ {Provider: "vercel-ai-gateway", ID: "zai/glm-5-turbo", DisplayName: "zai/glm-5-turbo", ContextWindow: 202800, MaxOutput: 131100, Reasoning: true, PriceInput: 1.2, PriceOutput: 4, PriceCacheRead: 0.24, BaseURL: "https://ai-gateway.vercel.sh"},
510+ {Provider: "vercel-ai-gateway", ID: "zai/glm-5.1", DisplayName: "zai/glm-5.1", ContextWindow: 202800, MaxOutput: 64000, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://ai-gateway.vercel.sh"},
511+ {Provider: "vercel-ai-gateway", ID: "zai/glm-5v-turbo", DisplayName: "zai/glm-5v-turbo", ContextWindow: 200000, MaxOutput: 128000, Reasoning: true, PriceInput: 1.2, PriceOutput: 4, PriceCacheRead: 0.24, BaseURL: "https://ai-gateway.vercel.sh"},
512+ // ----- xai -----
513+ {Provider: "xai", ID: "grok-3", DisplayName: "grok-3", ContextWindow: 131072, MaxOutput: 8192, Reasoning: false, PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.75, BaseURL: "https://api.x.ai/v1"},
514+ {Provider: "xai", ID: "grok-3-fast", DisplayName: "grok-3-fast", ContextWindow: 131072, MaxOutput: 8192, Reasoning: false, PriceInput: 5, PriceOutput: 25, PriceCacheRead: 1.25, BaseURL: "https://api.x.ai/v1"},
515+ {Provider: "xai", ID: "grok-4.20-0309-non-reasoning", DisplayName: "grok-4.20-0309-non-reasoning", ContextWindow: 2000000, MaxOutput: 30000, Reasoning: false, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.2, BaseURL: "https://api.x.ai/v1"},
516+ {Provider: "xai", ID: "grok-4.20-0309-reasoning", DisplayName: "grok-4.20-0309-reasoning", ContextWindow: 2000000, MaxOutput: 30000, Reasoning: true, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.2, BaseURL: "https://api.x.ai/v1"},
517+ {Provider: "xai", ID: "grok-4.3", DisplayName: "grok-4.3", ContextWindow: 1000000, MaxOutput: 30000, Reasoning: true, PriceInput: 1.25, PriceOutput: 2.5, PriceCacheRead: 0.2, BaseURL: "https://api.x.ai/v1"},
518+ {Provider: "xai", ID: "grok-build-0.1", DisplayName: "grok-build-0.1", ContextWindow: 256000, MaxOutput: 256000, Reasoning: true, PriceInput: 1, PriceOutput: 2, PriceCacheRead: 0.2, BaseURL: "https://api.x.ai/v1"},
519+ {Provider: "xai", ID: "grok-code-fast-1", DisplayName: "grok-code-fast-1", ContextWindow: 32768, MaxOutput: 8192, Reasoning: false, PriceInput: 0.2, PriceOutput: 1.5, PriceCacheRead: 0.02, BaseURL: "https://api.x.ai/v1"},
520+ // ----- xiaomi -----
521+ {Provider: "xiaomi", ID: "mimo-v2-flash", DisplayName: "mimo-v2-flash", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.3, PriceCacheRead: 0.01, BaseURL: "https://api.xiaomimimo.com/v1"},
522+ {Provider: "xiaomi", ID: "mimo-v2-omni", DisplayName: "mimo-v2-omni", ContextWindow: 262144, MaxOutput: 131072, Reasoning: true, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://api.xiaomimimo.com/v1"},
523+ {Provider: "xiaomi", ID: "mimo-v2-pro", DisplayName: "mimo-v2-pro", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.2, BaseURL: "https://api.xiaomimimo.com/v1"},
524+ {Provider: "xiaomi", ID: "mimo-v2.5", DisplayName: "mimo-v2.5", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://api.xiaomimimo.com/v1"},
525+ {Provider: "xiaomi", ID: "mimo-v2.5-pro", DisplayName: "mimo-v2.5-pro", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.2, BaseURL: "https://api.xiaomimimo.com/v1"},
526+ // ----- xiaomi-token-plan-ams -----
527+ {Provider: "xiaomi-token-plan-ams", ID: "mimo-v2-flash", DisplayName: "mimo-v2-flash", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.3, PriceCacheRead: 0.01, BaseURL: "https://token-plan-ams.xiaomimimo.com/v1"},
528+ {Provider: "xiaomi-token-plan-ams", ID: "mimo-v2-omni", DisplayName: "mimo-v2-omni", ContextWindow: 262144, MaxOutput: 131072, Reasoning: true, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://token-plan-ams.xiaomimimo.com/v1"},
529+ {Provider: "xiaomi-token-plan-ams", ID: "mimo-v2-pro", DisplayName: "mimo-v2-pro", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.2, BaseURL: "https://token-plan-ams.xiaomimimo.com/v1"},
530+ {Provider: "xiaomi-token-plan-ams", ID: "mimo-v2.5", DisplayName: "mimo-v2.5", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://token-plan-ams.xiaomimimo.com/v1"},
531+ {Provider: "xiaomi-token-plan-ams", ID: "mimo-v2.5-pro", DisplayName: "mimo-v2.5-pro", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.2, BaseURL: "https://token-plan-ams.xiaomimimo.com/v1"},
532+ // ----- xiaomi-token-plan-cn -----
533+ {Provider: "xiaomi-token-plan-cn", ID: "mimo-v2-flash", DisplayName: "mimo-v2-flash", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.3, PriceCacheRead: 0.01, BaseURL: "https://token-plan-cn.xiaomimimo.com/v1"},
534+ {Provider: "xiaomi-token-plan-cn", ID: "mimo-v2-omni", DisplayName: "mimo-v2-omni", ContextWindow: 262144, MaxOutput: 131072, Reasoning: true, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://token-plan-cn.xiaomimimo.com/v1"},
535+ {Provider: "xiaomi-token-plan-cn", ID: "mimo-v2-pro", DisplayName: "mimo-v2-pro", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.2, BaseURL: "https://token-plan-cn.xiaomimimo.com/v1"},
536+ {Provider: "xiaomi-token-plan-cn", ID: "mimo-v2.5", DisplayName: "mimo-v2.5", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://token-plan-cn.xiaomimimo.com/v1"},
537+ {Provider: "xiaomi-token-plan-cn", ID: "mimo-v2.5-pro", DisplayName: "mimo-v2.5-pro", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.2, BaseURL: "https://token-plan-cn.xiaomimimo.com/v1"},
538+ // ----- xiaomi-token-plan-sgp -----
539+ {Provider: "xiaomi-token-plan-sgp", ID: "mimo-v2-flash", DisplayName: "mimo-v2-flash", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.1, PriceOutput: 0.3, PriceCacheRead: 0.01, BaseURL: "https://token-plan-sgp.xiaomimimo.com/v1"},
540+ {Provider: "xiaomi-token-plan-sgp", ID: "mimo-v2-omni", DisplayName: "mimo-v2-omni", ContextWindow: 262144, MaxOutput: 131072, Reasoning: true, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://token-plan-sgp.xiaomimimo.com/v1"},
541+ {Provider: "xiaomi-token-plan-sgp", ID: "mimo-v2-pro", DisplayName: "mimo-v2-pro", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.2, BaseURL: "https://token-plan-sgp.xiaomimimo.com/v1"},
542+ {Provider: "xiaomi-token-plan-sgp", ID: "mimo-v2.5", DisplayName: "mimo-v2.5", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 0.4, PriceOutput: 2, PriceCacheRead: 0.08, BaseURL: "https://token-plan-sgp.xiaomimimo.com/v1"},
543+ {Provider: "xiaomi-token-plan-sgp", ID: "mimo-v2.5-pro", DisplayName: "mimo-v2.5-pro", ContextWindow: 1048576, MaxOutput: 131072, Reasoning: true, PriceInput: 1, PriceOutput: 3, PriceCacheRead: 0.2, BaseURL: "https://token-plan-sgp.xiaomimimo.com/v1"},
544+ // ----- zai -----
545+ {Provider: "zai", ID: "glm-4.5-air", DisplayName: "glm-4.5-air", ContextWindow: 131072, MaxOutput: 98304, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.z.ai/api/coding/paas/v4"},
546+ {Provider: "zai", ID: "glm-4.7", DisplayName: "glm-4.7", ContextWindow: 204800, MaxOutput: 131072, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.z.ai/api/coding/paas/v4"},
547+ {Provider: "zai", ID: "glm-5-turbo", DisplayName: "glm-5-turbo", ContextWindow: 200000, MaxOutput: 131072, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.z.ai/api/coding/paas/v4"},
548+ {Provider: "zai", ID: "glm-5.1", DisplayName: "glm-5.1", ContextWindow: 200000, MaxOutput: 131072, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.z.ai/api/coding/paas/v4"},
549+ {Provider: "zai", ID: "glm-5v-turbo", DisplayName: "glm-5v-turbo", ContextWindow: 200000, MaxOutput: 131072, Reasoning: true, PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, BaseURL: "https://api.z.ai/api/coding/paas/v4"},
550+}
A internal/provider/extra_models.go +41 -0
@@ -0,0 +1,41 @@
1+package provider
2+
3+// Hand-curated catalog supplements that the bulk builtin catalog
4+// (catalog_builtin.go) doesn't cover.
5+//
6+// - openai-responses: zot-specific provider id for the public OpenAI
7+// Responses API (api.openai.com/v1/responses). Separate from
8+// `openai` (Chat Completions) and `openai-codex` (ChatGPT subscription).
9+// - kimi/kimi-k2-thinking: matches the Anthropic-Messages endpoint
10+// at api.kimi.com/coding; not present in the bulk builtin file
11+// because the `kimi` provider lives in models.go alongside the
12+// other curated providers.
13+
14+func init() { Catalog = append(Catalog, supplementCatalog...) }
15+
16+var supplementCatalog = []Model{
17+ // ----- kimi (anthropic-messages on api.kimi.com/coding) -----
18+ {Provider: "kimi", ID: "kimi-k2-thinking", DisplayName: "Kimi K2 Thinking",
19+ ContextWindow: 262144, MaxOutput: 32000, Reasoning: true,
20+ BaseURL: "https://api.kimi.com/coding"},
21+
22+ // ----- openai-responses (public OpenAI Responses API) -----
23+ {Provider: "openai-responses", ID: "gpt-5", DisplayName: "GPT-5 (Responses)",
24+ ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
25+ PriceInput: 1.25, PriceOutput: 10.00, PriceCacheRead: 0.125},
26+ {Provider: "openai-responses", ID: "gpt-5-mini", DisplayName: "GPT-5 mini (Responses)",
27+ ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
28+ PriceInput: 0.25, PriceOutput: 2.00, PriceCacheRead: 0.025},
29+ {Provider: "openai-responses", ID: "gpt-5-nano", DisplayName: "GPT-5 nano (Responses)",
30+ ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
31+ PriceInput: 0.05, PriceOutput: 0.40, PriceCacheRead: 0.005},
32+ {Provider: "openai-responses", ID: "gpt-5-codex", DisplayName: "GPT-5 Codex (Responses)",
33+ ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
34+ PriceInput: 1.25, PriceOutput: 10.00, PriceCacheRead: 0.125},
35+ {Provider: "openai-responses", ID: "o3", DisplayName: "o3 (Responses)",
36+ ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
37+ PriceInput: 2.00, PriceOutput: 8.00, PriceCacheRead: 0.50},
38+ {Provider: "openai-responses", ID: "o4-mini", DisplayName: "o4-mini (Responses)",
39+ ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
40+ PriceInput: 1.10, PriceOutput: 4.40, PriceCacheRead: 0.275},
41+}
A internal/provider/extra_providers.go +363 -0
@@ -0,0 +1,363 @@
1+package provider
2+
3+// Extra third-party providers.
4+//
5+// Most are OpenAI Chat Completions–compatible, so they reuse `openaiClient`
6+// with a different name + base URL. A handful speak the Anthropic Messages
7+// API and reuse `anthropicClient` via NewAnthropicCompat below.
8+//
9+// Providers with a non-trivial protocol (Bedrock Converse, Vertex SSE, Azure
10+// Responses, Mistral Conversations) are stubbed so the host wiring compiles.
11+// Filling those in is its own work — Bedrock needs SigV4 + Converse stream
12+// parsing, Vertex needs ADC, Azure needs the Responses API shape, Mistral
13+// needs its bespoke Conversations API.
14+
15+import (
16+ "context"
17+ "fmt"
18+ "net/http"
19+ "os"
20+ "strings"
21+)
22+
23+// ----------------------------------------------------------------------
24+// OpenAI Chat Completions–compatible providers. All of these just route
25+// Chat Completions to a different base URL with a different name. The
26+// default Authorization: Bearer header from openaiClient is correct for
27+// every one of them.
28+// ----------------------------------------------------------------------
29+
30+// newOpenAICompat is the shared constructor for OpenAI-completions clones.
31+func newOpenAICompat(name, apiKey, baseURL, fallbackBaseURL string) Client {
32+ if baseURL == "" {
33+ baseURL = fallbackBaseURL
34+ }
35+ return &openaiClient{
36+ apiKey: apiKey,
37+ baseURL: strings.TrimRight(baseURL, "/"),
38+ name: name,
39+ http: &http.Client{Timeout: 0},
40+ }
41+}
42+
43+// NewMoonshot is the global Moonshot AI endpoint (Kimi-K2 family by id).
44+// Provider id is `moonshotai`.
45+func NewMoonshot(apiKey, baseURL string) Client {
46+ return newOpenAICompat("moonshotai", apiKey, baseURL, "https://api.moonshot.ai/v1")
47+}
48+
49+// NewMoonshotCN is the China-region Moonshot endpoint. Same model ids as
50+// the global flavor, different base URL.
51+func NewMoonshotCN(apiKey, baseURL string) Client {
52+ return newOpenAICompat("moonshotai-cn", apiKey, baseURL, "https://api.moonshot.cn/v1")
53+}
54+
55+// NewCerebras: ultra-fast inference (Llama/Qwen/GPT-OSS/GLM).
56+func NewCerebras(apiKey, baseURL string) Client {
57+ return newOpenAICompat("cerebras", apiKey, baseURL, "https://api.cerebras.ai/v1")
58+}
59+
60+// NewGroq: LPU inference (Llama/Kimi/Qwen/GPT-OSS).
61+func NewGroq(apiKey, baseURL string) Client {
62+ return newOpenAICompat("groq", apiKey, baseURL, "https://api.groq.com/openai/v1")
63+}
64+
65+// NewXAI: xAI Grok.
66+func NewXAI(apiKey, baseURL string) Client {
67+ return newOpenAICompat("xai", apiKey, baseURL, "https://api.x.ai/v1")
68+}
69+
70+// NewTogether: Together.ai aggregator.
71+func NewTogether(apiKey, baseURL string) Client {
72+ return newOpenAICompat("together", apiKey, baseURL, "https://api.together.ai/v1")
73+}
74+
75+// NewHuggingFace: HF inference router.
76+func NewHuggingFace(apiKey, baseURL string) Client {
77+ return newOpenAICompat("huggingface", apiKey, baseURL, "https://router.huggingface.co/v1")
78+}
79+
80+// NewZAI: Z.AI GLM family.
81+func NewZAI(apiKey, baseURL string) Client {
82+ return newOpenAICompat("zai", apiKey, baseURL, "https://api.z.ai/api/coding/paas/v4")
83+}
84+
85+// NewXiaomi: Xiaomi MiMo family (default endpoint).
86+func NewXiaomi(apiKey, baseURL string) Client {
87+ return newOpenAICompat("xiaomi", apiKey, baseURL, "https://api.xiaomimimo.com/v1")
88+}
89+
90+// NewXiaomiTokenPlan creates a regional Xiaomi token-plan client.
91+// region must be "ams", "cn", or "sgp", matching the three
92+// `xiaomi-token-plan-*` provider ids.
93+func NewXiaomiTokenPlan(region, apiKey, baseURL string) Client {
94+ var fallback string
95+ var name string
96+ switch region {
97+ case "ams":
98+ fallback = "https://token-plan-ams.xiaomimimo.com/v1"
99+ name = "xiaomi-token-plan-ams"
100+ case "cn":
101+ fallback = "https://token-plan-cn.xiaomimimo.com/v1"
102+ name = "xiaomi-token-plan-cn"
103+ case "sgp":
104+ fallback = "https://token-plan-sgp.xiaomimimo.com/v1"
105+ name = "xiaomi-token-plan-sgp"
106+ default:
107+ panic(fmt.Sprintf("xiaomi token-plan: unknown region %q", region))
108+ }
109+ return newOpenAICompat(name, apiKey, baseURL, fallback)
110+}
111+
112+// NewOpenRouter: OpenRouter aggregator. Unlocks dozens of upstream
113+// models with one key.
114+func NewOpenRouter(apiKey, baseURL string) Client {
115+ return newOpenAICompat("openrouter", apiKey, baseURL, "https://openrouter.ai/api/v1")
116+}
117+
118+// NewOpenCode is the opencode.ai Zen endpoint. Mixed APIs upstream; this
119+// constructor wires the openai-completions flavor only. Models that need
120+// the anthropic-messages flavor under the same provider should be built
121+// with NewAnthropicCompat against the same base URL.
122+func NewOpenCode(apiKey, baseURL string) Client {
123+ return newOpenAICompat("opencode", apiKey, baseURL, "https://opencode.ai/zen/v1")
124+}
125+
126+// NewOpenCodeGo is the opencode-go variant.
127+func NewOpenCodeGo(apiKey, baseURL string) Client {
128+ return newOpenAICompat("opencode-go", apiKey, baseURL, "https://opencode.ai/zen/go/v1")
129+}
130+
131+// NewMinimaxOpenAI is the OpenAI-completions flavor of MiniMax, in case
132+// downstream models switch from anthropic-messages. The main MiniMax route
133+// uses anthropic-messages; see NewMinimaxAnthropic below.
134+func NewMinimaxOpenAI(apiKey, baseURL string) Client {
135+ return newOpenAICompat("minimax", apiKey, baseURL, "https://api.minimax.io/v1")
136+}
137+
138+// NewMinimaxCNOpenAI is the CN-region MiniMax (openai-completions).
139+func NewMinimaxCNOpenAI(apiKey, baseURL string) Client {
140+ return newOpenAICompat("minimax-cn", apiKey, baseURL, "https://api.minimaxi.com/v1")
141+}
142+
143+// NewFireworksOpenAI is the OpenAI-completions flavor of Fireworks.
144+// The main route uses the anthropic-messages variant on
145+// api.fireworks.ai/inference; see NewFireworksAnthropic.
146+func NewFireworksOpenAI(apiKey, baseURL string) Client {
147+ return newOpenAICompat("fireworks", apiKey, baseURL, "https://api.fireworks.ai/inference/v1")
148+}
149+
150+// NewVercelGatewayOpenAI is Vercel AI Gateway's OpenAI-compat shim. The
151+// main route uses anthropic-messages; see NewVercelGatewayAnthropic.
152+func NewVercelGatewayOpenAI(apiKey, baseURL string) Client {
153+ return newOpenAICompat("vercel-ai-gateway", apiKey, baseURL, "https://ai-gateway.vercel.sh/v1")
154+}
155+
156+// ----------------------------------------------------------------------
157+// Anthropic Messages–compatible providers. These speak Anthropic's wire
158+// format but live behind a third party's base URL. They reuse
159+// anthropicClient with a custom name.
160+// ----------------------------------------------------------------------
161+
162+// NewAnthropicCompat returns an anthropicClient pinned to a non-default
163+// base URL and identifying as `name` for cost / logging purposes. Auth is
164+// API key (x-api-key header). For OAuth-fronted compatibles (rare) use
165+// NewAnthropicOAuth and rename via NameClient.
166+func NewAnthropicCompat(name, apiKey, baseURL string) Client {
167+ if baseURL == "" {
168+ baseURL = anthropicDefaultBaseURL
169+ }
170+ return &anthropicClient{
171+ apiKey: apiKey,
172+ baseURL: strings.TrimRight(baseURL, "/"),
173+ name: name,
174+ http: &http.Client{Timeout: 0},
175+ }
176+}
177+
178+// NewKimiCoding is the Kimi Code client: Kimi behind the Anthropic
179+// Messages API at https://api.kimi.com/coding. Replaces the older
180+// OpenAI-completions-on-/coding/v1 wiring.
181+func NewKimiCoding(apiKey, baseURL string) Client {
182+ return NewKimiCodingWithHeaders(apiKey, baseURL, nil)
183+}
184+
185+// NewKimiCodingWithHeaders is the headered variant used by OAuth.
186+func NewKimiCodingWithHeaders(apiKey, baseURL string, headers map[string]string) Client {
187+ if baseURL == "" {
188+ baseURL = "https://api.kimi.com/coding"
189+ }
190+ return &anthropicClient{
191+ apiKey: apiKey,
192+ baseURL: strings.TrimRight(baseURL, "/"),
193+ name: "kimi",
194+ headers: headers,
195+ http: &http.Client{Timeout: 0},
196+ }
197+}
198+
199+// NewMinimaxAnthropic is the anthropic-messages flavor on
200+// api.minimax.io/anthropic, catalogued under provider=minimax.
201+func NewMinimaxAnthropic(apiKey, baseURL string) Client {
202+ return NewAnthropicCompat("minimax", apiKey, firstNonEmptyString(baseURL, "https://api.minimax.io/anthropic"))
203+}
204+
205+// NewMinimaxCNAnthropic is the CN-region MiniMax (anthropic-messages).
206+func NewMinimaxCNAnthropic(apiKey, baseURL string) Client {
207+ return NewAnthropicCompat("minimax-cn", apiKey, firstNonEmptyString(baseURL, "https://api.minimaxi.com/anthropic"))
208+}
209+
210+// NewFireworksAnthropic is the main Fireworks route. The
211+// anthropic-messages-compatible endpoint at api.fireworks.ai/inference
212+// expects Anthropic-style request bodies; use this rather than the
213+// OpenAI flavor.
214+func NewFireworksAnthropic(apiKey, baseURL string) Client {
215+ return NewAnthropicCompat("fireworks", apiKey, firstNonEmptyString(baseURL, "https://api.fireworks.ai/inference"))
216+}
217+
218+// NewVercelGatewayAnthropic — Vercel AI Gateway anthropic-messages route.
219+func NewVercelGatewayAnthropic(apiKey, baseURL string) Client {
220+ return NewAnthropicCompat("vercel-ai-gateway", apiKey, firstNonEmptyString(baseURL, "https://ai-gateway.vercel.sh"))
221+}
222+
223+// ----------------------------------------------------------------------
224+// Stubbed providers — bigger protocol surface, follow-up work.
225+// ----------------------------------------------------------------------
226+
227+type unimplementedClient struct {
228+ name string
229+ hint string
230+}
231+
232+func (c *unimplementedClient) Name() string { return c.name }
233+
234+func (c *unimplementedClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
235+ return nil, fmt.Errorf("provider %q not yet implemented: %s", c.name, c.hint)
236+}
237+
238+// NewBedrock returns an AWS Bedrock client. See amazon_bedrock.go for the
239+// hand-rolled Converse-Stream wire-format parser. Auth is via
240+// AWS_BEARER_TOKEN_BEDROCK (the modern Bedrock API key flow); SigV4
241+// signing for IAM access-key + secret credentials is not yet wired.
242+func NewBedrock(apiKey, baseURL string) Client {
243+ return NewBedrockClient(apiKey, baseURL)
244+}
245+
246+// NewGoogleVertex returns a Vertex AI client. See google_vertex.go for
247+// the full auth + URL-rewrite implementation. Requires GOOGLE_CLOUD_PROJECT,
248+// GOOGLE_CLOUD_LOCATION, and either GOOGLE_CLOUD_API_KEY or a service-
249+// account JSON file pointed to by GOOGLE_APPLICATION_CREDENTIALS (or the
250+// default ADC location ~/.config/gcloud/application_default_credentials.json).
251+func NewGoogleVertex(apiKey, baseURL string) Client {
252+ return NewVertex(apiKey, baseURL)
253+}
254+
255+// NewAzureOpenAIResponses delegates to the real Azure OpenAI client.
256+// Despite the provider id mentioning "responses", zot uses Azure's
257+// Chat Completions endpoint (older but functionally complete for our
258+// agent loop) to avoid duplicating the full openai-responses wire
259+// client. Models register under provider id `azure-openai-responses`
260+// so user catalogs keep working unchanged.
261+func NewAzureOpenAIResponses(apiKey, baseURL string) Client {
262+ return NewAzureOpenAI(apiKey, baseURL)
263+}
264+
265+// NewMistral returns a Mistral client using their OpenAI-compatible Chat
266+// Completions endpoint at https://api.mistral.ai/v1. Mistral also offers a
267+// bespoke "Conversations" API, but the OpenAI-compat endpoint supports
268+// the same models with tool calling and streaming, so we use that for
269+// simplicity (no extra wire format to maintain).
270+func NewMistral(apiKey, baseURL string) Client {
271+ return newOpenAICompat("mistral", apiKey, baseURL, "https://api.mistral.ai/v1")
272+}
273+
274+// Cloudflare endpoints carry `{CLOUDFLARE_ACCOUNT_ID}` and (for the AI
275+// Gateway) `{CLOUDFLARE_GATEWAY_ID}` placeholders that we substitute
276+// from env vars at client-construction time. Workers AI uses standard
277+// `Authorization: Bearer <key>`; AI Gateway uses `cf-aig-authorization`
278+// (the upstream Authorization header is passed through to whichever
279+// downstream provider the gateway forwards to).
280+
281+func resolveCloudflareURL(template string) (string, error) {
282+ out := template
283+ for _, key := range []string{"CLOUDFLARE_ACCOUNT_ID", "CLOUDFLARE_GATEWAY_ID"} {
284+ placeholder := "{" + key + "}"
285+ if !strings.Contains(out, placeholder) {
286+ continue
287+ }
288+ v := os.Getenv(key)
289+ if v == "" {
290+ return "", fmt.Errorf("%s is required but not set in env", key)
291+ }
292+ out = strings.ReplaceAll(out, placeholder, v)
293+ }
294+ return out, nil
295+}
296+
297+// NewCloudflareWorkersAI returns an OpenAI-compatible client pinned to
298+// the Workers AI base URL with {CLOUDFLARE_ACCOUNT_ID} substituted.
299+// Returns an erroring client (deferred error on first Stream call) if
300+// the env var is missing, so the constructor signature stays the same.
301+func NewCloudflareWorkersAI(apiKey, baseURL string) Client {
302+ if baseURL == "" {
303+ baseURL = "https://api.cloudflare.com/client/v4/accounts/{CLOUDFLARE_ACCOUNT_ID}/ai/v1"
304+ }
305+ resolved, err := resolveCloudflareURL(baseURL)
306+ if err != nil {
307+ return &unimplementedClient{name: "cloudflare-workers-ai", hint: err.Error()}
308+ }
309+ return newOpenAICompat("cloudflare-workers-ai", apiKey, resolved, "")
310+}
311+
312+// NewCloudflareAIGateway returns the AI Gateway client (OpenAI compat
313+// route). Sends `cf-aig-authorization` instead of `Authorization` so
314+// the gateway authenticates the caller (downstream-provider auth is
315+// configured per-gateway in the Cloudflare dashboard).
316+func NewCloudflareAIGateway(apiKey, baseURL string) Client {
317+ if baseURL == "" {
318+ baseURL = "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/compat"
319+ }
320+ resolved, err := resolveCloudflareURL(baseURL)
321+ if err != nil {
322+ return &unimplementedClient{name: "cloudflare-ai-gateway", hint: err.Error()}
323+ }
324+ return &openaiClient{
325+ apiKey: apiKey,
326+ baseURL: strings.TrimRight(resolved, "/"),
327+ name: "cloudflare-ai-gateway",
328+ headers: map[string]string{"cf-aig-authorization": "Bearer " + apiKey},
329+ http: &http.Client{Timeout: 0},
330+ }
331+}
332+
333+// NewGithubCopilot returns a GitHub Copilot client. The provided
334+// credential must be a GitHub Personal Access Token (PAT) with Copilot
335+// access enabled; zot trades it for a short-lived Copilot token on
336+// every inference request (cached in memory until ~5min before expiry).
337+//
338+// Wire format: OpenAI Chat Completions. Copilot-specific headers
339+// (X-Initiator, Openai-Intent, Editor-Version, Editor-Plugin-Version,
340+// Copilot-Integration-Id, User-Agent) are added by the refresh
341+// transport. The model id passes through unchanged.
342+//
343+// baseURL is ignored: the canonical host is read from `proxy-ep=...` in
344+// the short-lived token's value.
345+func NewGithubCopilot(apiKey, _ string) Client {
346+ if apiKey == "" {
347+ return &unimplementedClient{name: "github-copilot", hint: "set COPILOT_GITHUB_TOKEN"}
348+ }
349+ return NewGithubCopilotClient(apiKey)
350+}
351+
352+// ----------------------------------------------------------------------
353+// helpers
354+// ----------------------------------------------------------------------
355+
356+func firstNonEmptyString(vals ...string) string {
357+ for _, v := range vals {
358+ if v != "" {
359+ return v
360+ }
361+ }
362+ return ""
363+}
M internal/provider/gemini.go +84 -11
@@ -8,8 +8,12 @@ import (
88 "fmt"
99 "io"
1010 "net/http"
11+ "os"
12+ "path/filepath"
1113 "strings"
1214 "time"
15+
16+ "github.com/google/uuid"
1317 )
1418
1519 // Google Gemini provider, talking directly to the Generative Language
@@ -145,8 +149,12 @@ func (c *geminiClient) buildRequest(req Request) (*gemRequest, string, error) {
145149 }
146150 }
147151
148- // Convert tool defs.
149- if len(req.Tools) > 0 {
152+ functionsEnabled := geminiSupportsFunctionCalling(m.ID)
153+
154+ // Convert tool defs. Gemini image-generation models reject function
155+ // declarations with "Function calling is not enabled for this model";
156+ // for those models, send a direct multimodal prompt instead.
157+ if functionsEnabled && len(req.Tools) > 0 {
150158 decls := make([]gemFunctionDecl, 0, len(req.Tools))
151159 for _, t := range req.Tools {
152160 schema := t.Schema
@@ -179,8 +187,13 @@ func (c *geminiClient) buildRequest(req Request) (*gemRequest, string, error) {
179187 }
180188 out.GenerationConfig = gc
181189
182- // Convert messages.
183- msgs := RepairOrphanedToolResults(req.Messages)
190+ // Convert messages. When function calling is disabled for the target
191+ // model, also remove historical functionCall/functionResponse parts;
192+ // image models should receive only text/image content.
193+ msgs := req.Messages
194+ if functionsEnabled {
195+ msgs = RepairOrphanedToolResults(req.Messages)
196+ }
184197 for _, msg := range msgs {
185198 switch msg.Role {
186199 case RoleUser:
@@ -190,12 +203,15 @@ func (c *geminiClient) buildRequest(req Request) (*gemRequest, string, error) {
190203 }
191204 out.Contents = append(out.Contents, gemContent{Role: "user", Parts: parts})
192205 case RoleAssistant:
193- parts := convertGemAssistantParts(msg.Content)
206+ parts := convertGemAssistantParts(msg.Content, functionsEnabled)
194207 if len(parts) == 0 {
195208 continue
196209 }
197210 out.Contents = append(out.Contents, gemContent{Role: "model", Parts: parts})
198211 case RoleTool:
212+ if !functionsEnabled {
213+ continue
214+ }
199215 // Each tool_result becomes a user-role message with a
200216 // functionResponse part. Gemini's protocol uses
201217 // "user" role for tool replies.
@@ -231,7 +247,7 @@ func convertGemUserParts(blocks []Content) []gemPart {
231247 return parts
232248 }
233249
234-func convertGemAssistantParts(blocks []Content) []gemPart {
250+func convertGemAssistantParts(blocks []Content, functionsEnabled bool) []gemPart {
235251 var parts []gemPart
236252 for _, b := range blocks {
237253 switch v := b.(type) {
@@ -241,6 +257,9 @@ func convertGemAssistantParts(blocks []Content) []gemPart {
241257 }
242258 parts = append(parts, gemPart{Text: v.Text})
243259 case ToolCallBlock:
260+ if !functionsEnabled {
261+ continue
262+ }
244263 args := v.Arguments
245264 if len(args) == 0 || !json.Valid(args) {
246265 args = json.RawMessage("{}")
@@ -256,6 +275,34 @@ func convertGemAssistantParts(blocks []Content) []gemPart {
256275 return parts
257276 }
258277
278+func geminiSupportsFunctionCalling(modelID string) bool {
279+ id := strings.ToLower(modelID)
280+ // Gemini image generation/editing models accept direct multimodal
281+ // prompts but reject tools/function declarations.
282+ if strings.Contains(id, "flash-image") || strings.Contains(id, "image-generation") {
283+ return false
284+ }
285+ return true
286+}
287+
288+func saveGeminiImageToWorkingDir(mimeType string, data []byte) (string, error) {
289+ ext := ".png"
290+ switch strings.ToLower(mimeType) {
291+ case "image/jpeg", "image/jpg":
292+ ext = ".jpg"
293+ case "image/webp":
294+ ext = ".webp"
295+ case "image/gif":
296+ ext = ".gif"
297+ }
298+ name := "zot-gemini-image-" + uuid.NewString() + ext
299+ path := filepath.Join(".", name)
300+ if err := os.WriteFile(path, data, 0o644); err != nil {
301+ return "", err
302+ }
303+ return path, nil
304+}
305+
259306 func convertGemToolResultParts(blocks []Content) []gemPart {
260307 var parts []gemPart
261308 for _, b := range blocks {
@@ -433,11 +480,13 @@ func (c *geminiClient) runStream(ctx context.Context, resp *http.Response, req R
433480 // candidate carries a list of parts (text or functionCall),
434481 // possibly accumulating across chunks.
435482 type blockEntry struct {
436- kind string // "text" | "tool_use"
437- textBuf strings.Builder
438- toolID string
439- toolName string
440- toolArgs strings.Builder
483+ kind string // "text" | "image" | "tool_use"
484+ textBuf strings.Builder
485+ image *ImageBlock
486+ imagePath string
487+ toolID string
488+ toolName string
489+ toolArgs strings.Builder
441490 }
442491 var (
443492 blocks []*blockEntry
@@ -456,6 +505,18 @@ func (c *geminiClient) runStream(ctx context.Context, resp *http.Response, req R
456505 currentText.textBuf.WriteString(delta)
457506 }
458507
508+ appendImage := func(mimeType, dataB64 string) {
509+ data, err := base64.StdEncoding.DecodeString(dataB64)
510+ if err != nil || len(data) == 0 {
511+ return
512+ }
513+ img := ImageBlock{MimeType: mimeType, Data: data}
514+ path, _ := saveGeminiImageToWorkingDir(mimeType, data)
515+ blocks = append(blocks, &blockEntry{kind: "image", image: &img, imagePath: path})
516+ // Image blocks break the current text run.
517+ currentText = nil
518+ }
519+
459520 startTool := func(name string, providedID string, args json.RawMessage) *blockEntry {
460521 toolCounter++
461522 id := providedID
@@ -484,6 +545,13 @@ func (c *geminiClient) runStream(ctx context.Context, resp *http.Response, req R
484545 if b.textBuf.Len() > 0 {
485546 content = append(content, TextBlock{Text: b.textBuf.String()})
486547 }
548+ case "image":
549+ if b.image != nil && len(b.image.Data) > 0 {
550+ content = append(content, *b.image)
551+ if b.imagePath != "" {
552+ content = append(content, TextBlock{Text: fmt.Sprintf("Saved image: `%s`", b.imagePath)})
553+ }
554+ }
487555 case "tool_use":
488556 args := b.toolArgs.String()
489557 if args == "" || !json.Valid([]byte(args)) {
@@ -524,6 +592,7 @@ func (c *geminiClient) runStream(ctx context.Context, resp *http.Response, req R
524592 Role string `json:"role"`
525593 Parts []struct {
526594 Text string `json:"text"`
595+ InlineData *gemInlineData `json:"inlineData"`
527596 Thought bool `json:"thought"`
528597 FunctionCall *gemFunctionCall `json:"functionCall"`
529598 FunctionCallID string `json:"id"`
@@ -556,6 +625,10 @@ func (c *geminiClient) runStream(ctx context.Context, resp *http.Response, req R
556625 }
557626 for _, cand := range chunk.Candidates {
558627 for _, part := range cand.Content.Parts {
628+ if part.InlineData != nil {
629+ appendImage(part.InlineData.MimeType, part.InlineData.Data)
630+ continue
631+ }
559632 if part.FunctionCall != nil {
560633 var args json.RawMessage
561634 if len(part.FunctionCall.Args) > 0 {
M internal/provider/gemini_test.go +79 -0
@@ -2,10 +2,12 @@ package provider
22
33 import (
44 "context"
5+ "encoding/base64"
56 "encoding/json"
67 "io"
78 "net/http"
89 "net/http/httptest"
10+ "os"
911 "strings"
1012 "testing"
1113 )
@@ -66,6 +68,51 @@ func TestGeminiStreamHappyPath(t *testing.T) {
6668 }
6769 }
6870
71+func TestGeminiStreamInlineImage(t *testing.T) {
72+ t.Chdir(t.TempDir())
73+ img := base64.StdEncoding.EncodeToString([]byte("png-bytes"))
74+ srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
75+ w.Header().Set("content-type", "text/event-stream")
76+ w.WriteHeader(http.StatusOK)
77+ _, _ = w.Write([]byte("data: " + `{"candidates":[{"content":{"role":"model","parts":[{"text":"here"},{"inlineData":{"mimeType":"image/png","data":"` + img + `"}}]},"finishReason":"STOP"}]}` + "\n\n"))
78+ }))
79+ defer srv.Close()
80+
81+ c := NewGemini("k", srv.URL)
82+ evs, err := c.Stream(context.Background(), Request{Model: "gemini-2.5-flash-image"})
83+ if err != nil {
84+ t.Fatal(err)
85+ }
86+ var done EventDone
87+ for ev := range evs {
88+ if e, ok := ev.(EventDone); ok {
89+ done = e
90+ }
91+ }
92+ if len(done.Message.Content) != 3 {
93+ t.Fatalf("content count=%d: %+v", len(done.Message.Content), done.Message.Content)
94+ }
95+ if tb, ok := done.Message.Content[0].(TextBlock); !ok || tb.Text != "here" {
96+ t.Fatalf("text block=%T %+v", done.Message.Content[0], done.Message.Content[0])
97+ }
98+ ib, ok := done.Message.Content[1].(ImageBlock)
99+ if !ok {
100+ t.Fatalf("want image block, got %T", done.Message.Content[1])
101+ }
102+ if ib.MimeType != "image/png" || string(ib.Data) != "png-bytes" {
103+ t.Fatalf("image=%q %q", ib.MimeType, string(ib.Data))
104+ }
105+ saved, ok := done.Message.Content[2].(TextBlock)
106+ if !ok || !strings.Contains(saved.Text, "zot-gemini-image-") || !strings.Contains(saved.Text, ".png") {
107+ t.Fatalf("saved path block=%T %+v", done.Message.Content[2], done.Message.Content[2])
108+ }
109+ path := strings.TrimPrefix(saved.Text, "Saved image: `")
110+ path = strings.TrimSuffix(path, "`")
111+ if _, err := os.Stat(path); err != nil {
112+ t.Fatalf("saved image missing at %q: %v", path, err)
113+ }
114+}
115+
69116 // TestGeminiToolCall covers the tool-call branch: a single
70117 // functionCall part should produce ToolStart/Args/End and the final
71118 // stop reason should be StopToolUse.
@@ -149,6 +196,38 @@ func TestGeminiBuildRequestSystemAndTools(t *testing.T) {
149196 }
150197 }
151198
199+// TestGeminiBuildRequestImageModelOmitsTools confirms image-generation
200+// models receive direct multimodal prompts without function declarations.
201+func TestGeminiBuildRequestImageModelOmitsTools(t *testing.T) {
202+ c := NewGemini("k", "https://example.invalid").(*geminiClient)
203+ wire, _, err := c.buildRequest(Request{
204+ Model: "gemini-2.5-flash-image",
205+ Tools: []Tool{
206+ {Name: "read", Description: "read a file", Schema: json.RawMessage(`{"type":"object","properties":{"path":{"type":"string"}}}`)},
207+ },
208+ Messages: []Message{
209+ {Role: RoleUser, Content: []Content{TextBlock{Text: "edit this image"}, ImageBlock{MimeType: "image/png", Data: []byte("png")}}},
210+ {Role: RoleAssistant, Content: []Content{TextBlock{Text: "checking"}, ToolCallBlock{ID: "1", Name: "read", Arguments: json.RawMessage(`{"path":"x"}`)}}},
211+ {Role: RoleTool, Content: []Content{ToolResultBlock{CallID: "1", Content: []Content{TextBlock{Text: "tool output"}}}}},
212+ },
213+ })
214+ if err != nil {
215+ t.Fatal(err)
216+ }
217+ if len(wire.Tools) != 0 {
218+ t.Fatalf("image model should omit tools, got %+v", wire.Tools)
219+ }
220+ if len(wire.Contents) != 2 {
221+ t.Fatalf("contents count=%d: %+v", len(wire.Contents), wire.Contents)
222+ }
223+ if len(wire.Contents[0].Parts) != 2 || wire.Contents[0].Parts[1].InlineData == nil {
224+ t.Fatalf("user image parts not preserved: %+v", wire.Contents[0].Parts)
225+ }
226+ if len(wire.Contents[1].Parts) != 1 || wire.Contents[1].Parts[0].FunctionCall != nil {
227+ t.Fatalf("assistant tool call not stripped: %+v", wire.Contents[1].Parts)
228+ }
229+}
230+
152231 // TestGeminiErrorStatus confirms HTTP error bodies bubble up.
153232 func TestGeminiErrorStatus(t *testing.T) {
154233 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
A internal/provider/github_copilot.go +198 -0
@@ -0,0 +1,198 @@
1+package provider
2+
3+// GitHub Copilot client.
4+//
5+// Auth chain:
6+//
7+// 1. User stores a GitHub Personal Access Token (PAT) in
8+// COPILOT_GITHUB_TOKEN (or via zot's auth.json, when wired). The PAT
9+// is the long-lived credential; it never hits the inference endpoint.
10+// 2. We trade the PAT for a short-lived Copilot token by hitting
11+// `GET https://api.github.com/copilot_internal/v2/token` with the PAT
12+// as `Authorization: Bearer` plus the Copilot identity headers
13+// (Editor-Version, Editor-Plugin-Version, Copilot-Integration-Id,
14+// User-Agent). The response carries `{ "token": "...", "expires_at": <unix> }`.
15+// 3. The short-lived token's value embeds a `proxy-ep=<host>` field that
16+// tells us the real API host (individual users: api.individual.githubcopilot.com).
17+// 4. Inference requests go to `<host>/chat/completions` with the
18+// short-lived token in `Authorization: Bearer` plus extras:
19+// - X-Initiator: user|agent
20+// - Openai-Intent: conversation-edits
21+// - Copilot-Vision-Request: true (when images present; not wired
22+// here because the zot openai client currently sends images inline)
23+//
24+// Token caching: short-lived tokens last ~30min. We cache one per PAT in
25+// memory for the process lifetime and refresh on demand. No disk cache.
26+
27+import (
28+ "context"
29+ "encoding/json"
30+ "fmt"
31+ "io"
32+ "net/http"
33+ "net/url"
34+ "strings"
35+ "sync"
36+ "time"
37+)
38+
39+// Copilot identity headers — these must match what the VS Code Copilot
40+// extension sends or the proxy rejects with 401.
41+var copilotIdentityHeaders = map[string]string{
42+ "User-Agent": "GitHubCopilotChat/0.35.0",
43+ "Editor-Version": "vscode/1.107.0",
44+ "Editor-Plugin-Version": "copilot-chat/0.35.0",
45+ "Copilot-Integration-Id": "vscode-chat",
46+}
47+
48+type copilotToken struct {
49+ value string
50+ expiresAt time.Time
51+ baseURL string
52+}
53+
54+// copilotTokenCache is a process-wide cache of short-lived Copilot tokens
55+// keyed by the user's PAT. Concurrency-safe; safe to call from multiple
56+// goroutines (a single agent loop is sequential, but extension intercepts
57+// can run in parallel).
58+type copilotTokenCache struct {
59+ mu sync.Mutex
60+ tokens map[string]copilotToken
61+ http *http.Client
62+}
63+
64+var copilotCache = &copilotTokenCache{
65+ tokens: map[string]copilotToken{},
66+ http: &http.Client{Timeout: 30 * time.Second},
67+}
68+
69+// exchange swaps a PAT for a short-lived Copilot token. Caller is
70+// expected to hold no lock; we acquire/release internally.
71+func (c *copilotTokenCache) exchange(ctx context.Context, pat string) (copilotToken, error) {
72+ url := "https://api.github.com/copilot_internal/v2/token"
73+ req, err := http.NewRequestWithContext(ctx, "GET", url, nil)
74+ if err != nil {
75+ return copilotToken{}, err
76+ }
77+ req.Header.Set("Authorization", "Bearer "+pat)
78+ req.Header.Set("Accept", "application/json")
79+ for k, v := range copilotIdentityHeaders {
80+ req.Header.Set(k, v)
81+ }
82+ resp, err := c.http.Do(req)
83+ if err != nil {
84+ return copilotToken{}, fmt.Errorf("copilot token exchange: %w", err)
85+ }
86+ defer resp.Body.Close()
87+ body, _ := io.ReadAll(resp.Body)
88+ if resp.StatusCode != http.StatusOK {
89+ return copilotToken{}, fmt.Errorf("copilot token exchange: http %d: %s", resp.StatusCode, strings.TrimSpace(string(body)))
90+ }
91+ var out struct {
92+ Token string `json:"token"`
93+ ExpiresAt int64 `json:"expires_at"`
94+ }
95+ if err := json.Unmarshal(body, &out); err != nil {
96+ return copilotToken{}, fmt.Errorf("copilot token: parse: %w", err)
97+ }
98+ if out.Token == "" {
99+ return copilotToken{}, fmt.Errorf("copilot token: empty")
100+ }
101+ base := parseCopilotProxyEndpoint(out.Token)
102+ if base == "" {
103+ base = "https://api.individual.githubcopilot.com"
104+ }
105+ exp := time.Unix(out.ExpiresAt, 0)
106+ if exp.IsZero() || time.Until(exp) < 0 {
107+ // Reference subtracts a 5min safety margin; we do the same so a
108+ // long-running tool call doesn't blow past expiry mid-flight.
109+ exp = time.Now().Add(25 * time.Minute)
110+ } else {
111+ exp = exp.Add(-5 * time.Minute)
112+ }
113+ return copilotToken{value: out.Token, expiresAt: exp, baseURL: base}, nil
114+}
115+
116+// get returns a fresh-enough Copilot token, refreshing if needed.
117+func (c *copilotTokenCache) get(ctx context.Context, pat string) (copilotToken, error) {
118+ c.mu.Lock()
119+ tok, ok := c.tokens[pat]
120+ c.mu.Unlock()
121+ if ok && time.Now().Before(tok.expiresAt) {
122+ return tok, nil
123+ }
124+ fresh, err := c.exchange(ctx, pat)
125+ if err != nil {
126+ return copilotToken{}, err
127+ }
128+ c.mu.Lock()
129+ c.tokens[pat] = fresh
130+ c.mu.Unlock()
131+ return fresh, nil
132+}
133+
134+// parseCopilotProxyEndpoint extracts `proxy-ep=<host>` from a Copilot
135+// short-lived token's value (it's not a JWT — it's an ad-hoc
136+// semicolon-separated key=value string) and converts proxy.* -> api.*.
137+// Returns "" if not found.
138+func parseCopilotProxyEndpoint(token string) string {
139+ for _, part := range strings.Split(token, ";") {
140+ if strings.HasPrefix(part, "proxy-ep=") {
141+ host := strings.TrimPrefix(part, "proxy-ep=")
142+ host = strings.TrimPrefix(host, "proxy.")
143+ return "https://api." + host
144+ }
145+ }
146+ return ""
147+}
148+
149+// copilotRefreshTransport wraps the default HTTP transport so every
150+// outgoing inference request gets the latest short-lived Copilot token
151+// in Authorization. Token exchange uses the PAT carried in the
152+// transport itself, not the request.
153+type copilotRefreshTransport struct {
154+ inner http.RoundTripper
155+ pat string
156+}
157+
158+func (t *copilotRefreshTransport) RoundTrip(req *http.Request) (*http.Response, error) {
159+ tok, err := copilotCache.get(req.Context(), t.pat)
160+ if err != nil {
161+ return nil, err
162+ }
163+ clone := req.Clone(req.Context())
164+ clone.Header.Set("Authorization", "Bearer "+tok.value)
165+ // Identity headers also required on inference requests.
166+ for k, v := range copilotIdentityHeaders {
167+ clone.Header.Set(k, v)
168+ }
169+ clone.Header.Set("X-Initiator", "agent")
170+ clone.Header.Set("Openai-Intent", "conversation-edits")
171+ // If the request URL host doesn't match the token's proxy-ep, rewrite
172+ // it. The openaiClient pinned a static host at construction time, but
173+ // the canonical host comes from the token.
174+ if tok.baseURL != "" {
175+ if u, err := url.Parse(tok.baseURL); err == nil && u.Host != clone.URL.Host {
176+ clone.URL.Scheme = u.Scheme
177+ clone.URL.Host = u.Host
178+ }
179+ }
180+ return t.inner.RoundTrip(clone)
181+}
182+
183+// NewGithubCopilotClient returns a Copilot-pinned OpenAI-compat client.
184+// The pat must be a GitHub Personal Access Token with Copilot access.
185+func NewGithubCopilotClient(pat string) Client {
186+ httpClient := &http.Client{
187+ Transport: &copilotRefreshTransport{inner: http.DefaultTransport, pat: pat},
188+ Timeout: 0,
189+ }
190+ // Initial baseURL is a sane default; copilotRefreshTransport rewrites
191+ // the host on every request based on the freshly-issued token.
192+ return &openaiClient{
193+ apiKey: pat, // unused at the wire level (transport overrides Auth) but kept for parity
194+ baseURL: "https://api.individual.githubcopilot.com",
195+ name: "github-copilot",
196+ http: httpClient,
197+ }
198+}
A internal/provider/google_vertex.go +362 -0
@@ -0,0 +1,362 @@
1+package provider
2+
3+// Google Vertex AI client.
4+//
5+// Wire format is identical to the public Generative Language API
6+// (generateContent / streamGenerateContent), so we reuse the existing
7+// geminiClient and rewrite the outgoing URL + auth via a RoundTripper.
8+//
9+// URL shape:
10+//
11+// POST https://{location}-aiplatform.googleapis.com/v1/projects/{project}/
12+// locations/{location}/publishers/google/models/{model}:streamGenerateContent?alt=sse
13+//
14+// Auth: two paths supported.
15+//
16+// - **API key** (GOOGLE_CLOUD_API_KEY): `x-goog-api-key` header. Simplest
17+// setup; works for users who have created a Vertex AI API key in the
18+// GCP console. We use this directly, no token exchange.
19+// - **Service-account JSON** pointed to by GOOGLE_APPLICATION_CREDENTIALS:
20+// we read the JSON, build a JWT signed with RS256 using the private
21+// key, exchange it at `https://oauth2.googleapis.com/token` for a
22+// short-lived (1h) access token, cache it in memory, and refresh on
23+// demand. Sent as `Authorization: Bearer <token>`.
24+// - **gcloud user OAuth** (~/.config/gcloud/application_default_credentials.json
25+// with `type: "authorized_user"`): we read the JSON, pull the
26+// client_id/client_secret/refresh_token, exchange the refresh token
27+// at `https://oauth2.googleapis.com/token` for an access token,
28+// same cache + refresh logic as service-account.
29+//
30+// Config env vars (read at construction time):
31+// GOOGLE_CLOUD_PROJECT required
32+// GOOGLE_CLOUD_LOCATION required (e.g. us-central1)
33+// GOOGLE_CLOUD_API_KEY optional; preferred over service-account
34+// GOOGLE_APPLICATION_CREDENTIALS optional; service-account JSON path
35+
36+import (
37+ "context"
38+ "crypto"
39+ "crypto/rand"
40+ "crypto/rsa"
41+ "crypto/sha256"
42+ "crypto/x509"
43+ "encoding/base64"
44+ "encoding/json"
45+ "encoding/pem"
46+ "fmt"
47+ "io"
48+ "net/http"
49+ "net/url"
50+ "os"
51+ "strings"
52+ "sync"
53+ "time"
54+)
55+
56+type vertexConfig struct {
57+ project string
58+ location string
59+ apiKey string // empty if using OAuth-style auth
60+
61+ // Auth mode. Exactly one of (saEmail+saPrivateKey) or
62+ // (userClientID+userClientSecret+userRefreshToken) is populated when
63+ // apiKey is empty.
64+ saEmail string
65+ saPrivateKey *rsa.PrivateKey
66+ saTokenURI string
67+
68+ userClientID string
69+ userClientSecret string
70+ userRefreshToken string
71+ userTokenURI string
72+}
73+
74+// cacheKey identifies the credential set for the token cache. We key on
75+// the auth principal (SA email or user client_id) so multiple configs
76+// in the same process don't trample each other.
77+func (c *vertexConfig) cacheKey() string {
78+ if c.saEmail != "" {
79+ return "sa:" + c.saEmail
80+ }
81+ return "user:" + c.userClientID
82+}
83+
84+func loadVertexConfig() (*vertexConfig, error) {
85+ cfg := &vertexConfig{
86+ project: os.Getenv("GOOGLE_CLOUD_PROJECT"),
87+ location: os.Getenv("GOOGLE_CLOUD_LOCATION"),
88+ apiKey: os.Getenv("GOOGLE_CLOUD_API_KEY"),
89+ }
90+ if cfg.location == "" {
91+ cfg.location = "us-central1"
92+ }
93+ if cfg.project == "" {
94+ return nil, fmt.Errorf("vertex: GOOGLE_CLOUD_PROJECT not set")
95+ }
96+ if cfg.apiKey != "" {
97+ return cfg, nil
98+ }
99+ // Try service-account JSON.
100+ credPath := os.Getenv("GOOGLE_APPLICATION_CREDENTIALS")
101+ if credPath == "" {
102+ // ADC default path. Mirrors `gcloud auth application-default login`.
103+ if home, err := os.UserHomeDir(); err == nil {
104+ candidate := home + "/.config/gcloud/application_default_credentials.json"
105+ if _, err := os.Stat(candidate); err == nil {
106+ credPath = candidate
107+ }
108+ }
109+ }
110+ if credPath == "" {
111+ return nil, fmt.Errorf("vertex: no auth — set GOOGLE_CLOUD_API_KEY or GOOGLE_APPLICATION_CREDENTIALS")
112+ }
113+ b, err := os.ReadFile(credPath)
114+ if err != nil {
115+ return nil, fmt.Errorf("vertex: read credentials %q: %w", credPath, err)
116+ }
117+ var raw struct {
118+ Type string `json:"type"`
119+ ClientEmail string `json:"client_email"`
120+ PrivateKey string `json:"private_key"`
121+ TokenURI string `json:"token_uri"`
122+ ClientID string `json:"client_id"`
123+ ClientSecret string `json:"client_secret"`
124+ RefreshToken string `json:"refresh_token"`
125+ }
126+ if err := json.Unmarshal(b, &raw); err != nil {
127+ return nil, fmt.Errorf("vertex: parse credentials: %w", err)
128+ }
129+ switch raw.Type {
130+ case "service_account":
131+ if raw.ClientEmail == "" || raw.PrivateKey == "" {
132+ return nil, fmt.Errorf("vertex: service_account JSON missing client_email or private_key")
133+ }
134+ block, _ := pem.Decode([]byte(raw.PrivateKey))
135+ if block == nil {
136+ return nil, fmt.Errorf("vertex: private_key is not PEM-encoded")
137+ }
138+ key, err := x509.ParsePKCS8PrivateKey(block.Bytes)
139+ if err != nil {
140+ // Try PKCS#1 as a fallback (older keys).
141+ k1, err2 := x509.ParsePKCS1PrivateKey(block.Bytes)
142+ if err2 != nil {
143+ return nil, fmt.Errorf("vertex: parse private_key: %w", err)
144+ }
145+ cfg.saPrivateKey = k1
146+ } else {
147+ rsaKey, ok := key.(*rsa.PrivateKey)
148+ if !ok {
149+ return nil, fmt.Errorf("vertex: private_key is not RSA")
150+ }
151+ cfg.saPrivateKey = rsaKey
152+ }
153+ cfg.saEmail = raw.ClientEmail
154+ cfg.saTokenURI = raw.TokenURI
155+ if cfg.saTokenURI == "" {
156+ cfg.saTokenURI = "https://oauth2.googleapis.com/token"
157+ }
158+ case "authorized_user":
159+ // gcloud auth application-default login produces this shape.
160+ if raw.ClientID == "" || raw.ClientSecret == "" || raw.RefreshToken == "" {
161+ return nil, fmt.Errorf("vertex: authorized_user JSON missing client_id/client_secret/refresh_token")
162+ }
163+ cfg.userClientID = raw.ClientID
164+ cfg.userClientSecret = raw.ClientSecret
165+ cfg.userRefreshToken = raw.RefreshToken
166+ cfg.userTokenURI = raw.TokenURI
167+ if cfg.userTokenURI == "" {
168+ cfg.userTokenURI = "https://oauth2.googleapis.com/token"
169+ }
170+ default:
171+ return nil, fmt.Errorf("vertex: unsupported credentials type %q (expected service_account or authorized_user)", raw.Type)
172+ }
173+ return cfg, nil
174+}
175+
176+// vertexTokenCache caches one access token per service-account email.
177+type vertexTokenCache struct {
178+ mu sync.Mutex
179+ tokens map[string]struct {
180+ value string
181+ expiresAt time.Time
182+ }
183+ http *http.Client
184+}
185+
186+var vertexCache = &vertexTokenCache{
187+ tokens: map[string]struct {
188+ value string
189+ expiresAt time.Time
190+ }{},
191+ http: &http.Client{Timeout: 30 * time.Second},
192+}
193+
194+func b64url(b []byte) string { return base64.RawURLEncoding.EncodeToString(b) }
195+
196+// signServiceAccountJWT builds and signs a JWT for the
197+// `urn:ietf:params:oauth:grant-type:jwt-bearer` token exchange.
198+func signServiceAccountJWT(cfg *vertexConfig) (string, error) {
199+ header := map[string]string{"alg": "RS256", "typ": "JWT"}
200+ now := time.Now().Unix()
201+ claims := map[string]any{
202+ "iss": cfg.saEmail,
203+ "scope": "https://www.googleapis.com/auth/cloud-platform",
204+ "aud": cfg.saTokenURI,
205+ "exp": now + 3600,
206+ "iat": now,
207+ }
208+ hb, _ := json.Marshal(header)
209+ cb, _ := json.Marshal(claims)
210+ signingInput := b64url(hb) + "." + b64url(cb)
211+ hashed := sha256.Sum256([]byte(signingInput))
212+ sig, err := rsa.SignPKCS1v15(rand.Reader, cfg.saPrivateKey, crypto.SHA256, hashed[:])
213+ if err != nil {
214+ return "", err
215+ }
216+ return signingInput + "." + b64url(sig), nil
217+}
218+
219+func (c *vertexTokenCache) get(ctx context.Context, cfg *vertexConfig) (string, error) {
220+ key := cfg.cacheKey()
221+ c.mu.Lock()
222+ tok, ok := c.tokens[key]
223+ c.mu.Unlock()
224+ if ok && time.Now().Before(tok.expiresAt) {
225+ return tok.value, nil
226+ }
227+
228+ var form url.Values
229+ var endpoint string
230+ switch {
231+ case cfg.saPrivateKey != nil:
232+ jwt, err := signServiceAccountJWT(cfg)
233+ if err != nil {
234+ return "", fmt.Errorf("vertex: sign jwt: %w", err)
235+ }
236+ form = url.Values{
237+ "grant_type": {"urn:ietf:params:oauth:grant-type:jwt-bearer"},
238+ "assertion": {jwt},
239+ }
240+ endpoint = cfg.saTokenURI
241+ case cfg.userRefreshToken != "":
242+ form = url.Values{
243+ "grant_type": {"refresh_token"},
244+ "client_id": {cfg.userClientID},
245+ "client_secret": {cfg.userClientSecret},
246+ "refresh_token": {cfg.userRefreshToken},
247+ }
248+ endpoint = cfg.userTokenURI
249+ default:
250+ return "", fmt.Errorf("vertex: no usable auth (need service_account or authorized_user creds)")
251+ }
252+
253+ req, err := http.NewRequestWithContext(ctx, "POST", endpoint, strings.NewReader(form.Encode()))
254+ if err != nil {
255+ return "", err
256+ }
257+ req.Header.Set("content-type", "application/x-www-form-urlencoded")
258+ resp, err := c.http.Do(req)
259+ if err != nil {
260+ return "", fmt.Errorf("vertex: token exchange: %w", err)
261+ }
262+ defer resp.Body.Close()
263+ body, _ := io.ReadAll(resp.Body)
264+ if resp.StatusCode != http.StatusOK {
265+ return "", fmt.Errorf("vertex: token exchange http %d: %s", resp.StatusCode, strings.TrimSpace(string(body)))
266+ }
267+ var out struct {
268+ AccessToken string `json:"access_token"`
269+ ExpiresIn int `json:"expires_in"`
270+ }
271+ if err := json.Unmarshal(body, &out); err != nil {
272+ return "", fmt.Errorf("vertex: parse token: %w", err)
273+ }
274+ if out.AccessToken == "" {
275+ return "", fmt.Errorf("vertex: empty access token")
276+ }
277+ exp := time.Now().Add(time.Duration(out.ExpiresIn-60) * time.Second)
278+ c.mu.Lock()
279+ c.tokens[key] = struct {
280+ value string
281+ expiresAt time.Time
282+ }{out.AccessToken, exp}
283+ c.mu.Unlock()
284+ return out.AccessToken, nil
285+}
286+
287+// vertexTransport rewrites the outgoing Gemini URL into the Vertex shape
288+// and replaces the x-goog-api-key header with either another
289+// x-goog-api-key (for the API-key flow) or Authorization: Bearer (for
290+// the service-account flow).
291+type vertexTransport struct {
292+ inner http.RoundTripper
293+ cfg *vertexConfig
294+}
295+
296+func (t *vertexTransport) RoundTrip(req *http.Request) (*http.Response, error) {
297+ // Path looks like /v1beta/models/{model}:streamGenerateContent. Pull
298+ // out the model id and rewrite.
299+ const prefix = "/v1beta/models/"
300+ if !strings.HasPrefix(req.URL.Path, prefix) {
301+ return nil, fmt.Errorf("vertex: unexpected request path %q", req.URL.Path)
302+ }
303+ rest := strings.TrimPrefix(req.URL.Path, prefix)
304+ // rest = "<model>:streamGenerateContent"
305+ sep := strings.IndexByte(rest, ':')
306+ if sep < 0 {
307+ return nil, fmt.Errorf("vertex: malformed request path %q", req.URL.Path)
308+ }
309+ modelID := rest[:sep]
310+ verb := rest[sep:] // ":streamGenerateContent"
311+ newPath := fmt.Sprintf("/v1/projects/%s/locations/%s/publishers/google/models/%s%s",
312+ t.cfg.project, t.cfg.location, modelID, verb)
313+ clone := req.Clone(req.Context())
314+ clone.URL.Host = t.cfg.location + "-aiplatform.googleapis.com"
315+ clone.URL.Path = newPath
316+ // Drop existing auth header and pick the right one.
317+ clone.Header.Del("x-goog-api-key")
318+ if t.cfg.apiKey != "" {
319+ clone.Header.Set("x-goog-api-key", t.cfg.apiKey)
320+ } else {
321+ tok, err := vertexCache.get(req.Context(), t.cfg)
322+ if err != nil {
323+ return nil, err
324+ }
325+ clone.Header.Set("Authorization", "Bearer "+tok)
326+ }
327+ return t.inner.RoundTrip(clone)
328+}
329+
330+// NewVertex returns a Vertex AI client. The apiKey argument is ignored
331+// in favor of env-based config (GOOGLE_CLOUD_API_KEY or
332+// GOOGLE_APPLICATION_CREDENTIALS), since Vertex's auth model doesn't fit
333+// the "just paste a key" interface other providers use.
334+func NewVertex(_ string, _ string) Client {
335+ cfg, err := loadVertexConfig()
336+ if err != nil {
337+ return &unimplementedClient{name: "google-vertex", hint: err.Error()}
338+ }
339+ inner := &geminiClient{
340+ apiKey: "vertex-placeholder", // overwritten by transport
341+ baseURL: "https://" + cfg.location + "-aiplatform.googleapis.com",
342+ http: &http.Client{
343+ Transport: &vertexTransport{inner: http.DefaultTransport, cfg: cfg},
344+ Timeout: 0,
345+ },
346+ }
347+ // Wrap so Name() reports "google-vertex" instead of "google".
348+ return &renamedClient{inner: inner, name: "google-vertex"}
349+}
350+
351+// renamedClient wraps a Client and overrides its Name(). Used for
352+// provider-id remapping (Vertex reuses gemini's wire client but should
353+// report as google-vertex for cost / cache / log purposes).
354+type renamedClient struct {
355+ inner Client
356+ name string
357+}
358+
359+func (r *renamedClient) Name() string { return r.name }
360+func (r *renamedClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
361+ return r.inner.Stream(ctx, req)
362+}
A internal/provider/google_vertex_test.go +57 -0
@@ -0,0 +1,57 @@
1+package provider
2+
3+import (
4+ "os"
5+ "path/filepath"
6+ "testing"
7+)
8+
9+func TestVertexConfigParsesAuthorizedUser(t *testing.T) {
10+ // Write a fake ADC user-OAuth file and point GOOGLE_APPLICATION_CREDENTIALS at it.
11+ tmp := t.TempDir()
12+ path := filepath.Join(tmp, "adc.json")
13+ body := `{
14+ "type": "authorized_user",
15+ "client_id": "fake.apps.googleusercontent.com",
16+ "client_secret": "fake-secret",
17+ "refresh_token": "1//fake-refresh"
18+ }`
19+ if err := os.WriteFile(path, []byte(body), 0o600); err != nil {
20+ t.Fatal(err)
21+ }
22+ t.Setenv("GOOGLE_APPLICATION_CREDENTIALS", path)
23+ t.Setenv("GOOGLE_CLOUD_PROJECT", "test-proj")
24+ t.Setenv("GOOGLE_CLOUD_API_KEY", "") // ensure SA / user path is chosen
25+ cfg, err := loadVertexConfig()
26+ if err != nil {
27+ t.Fatalf("loadVertexConfig: %v", err)
28+ }
29+ if cfg.userClientID != "fake.apps.googleusercontent.com" {
30+ t.Errorf("client_id = %q", cfg.userClientID)
31+ }
32+ if cfg.userRefreshToken != "1//fake-refresh" {
33+ t.Errorf("refresh_token = %q", cfg.userRefreshToken)
34+ }
35+ if cfg.userTokenURI != "https://oauth2.googleapis.com/token" {
36+ t.Errorf("token_uri default not applied: %q", cfg.userTokenURI)
37+ }
38+ if cfg.cacheKey() != "user:fake.apps.googleusercontent.com" {
39+ t.Errorf("cacheKey = %q", cfg.cacheKey())
40+ }
41+}
42+
43+func TestVertexConfigRejectsBadType(t *testing.T) {
44+ tmp := t.TempDir()
45+ path := filepath.Join(tmp, "adc.json")
46+ body := `{"type": "something_else"}`
47+ if err := os.WriteFile(path, []byte(body), 0o600); err != nil {
48+ t.Fatal(err)
49+ }
50+ t.Setenv("GOOGLE_APPLICATION_CREDENTIALS", path)
51+ t.Setenv("GOOGLE_CLOUD_PROJECT", "test-proj")
52+ t.Setenv("GOOGLE_CLOUD_API_KEY", "")
53+ _, err := loadVertexConfig()
54+ if err == nil {
55+ t.Fatal("expected error for unknown credential type")
56+ }
57+}
M internal/provider/models.go +3 -4
@@ -111,14 +111,13 @@ var Catalog = []Model{
111111 },
112112
113113 // ---- Kimi / Kimi Code ----
114+ // Anthropic-messages on https://api.kimi.com/coding (no /v1 suffix;
115+ // the Anthropic client appends /v1/messages itself).
114116 {
115117 Provider: "kimi", ID: "kimi-for-coding", DisplayName: "Kimi-k2.6",
116118 ContextWindow: 262144, MaxOutput: 32000, Reasoning: true,
117- // Kimi Coding's public model metadata currently reports zero-priced
118- // usage for this endpoint; keep the cost meter explicit instead of
119- // relying on the struct's zero defaults.
120119 PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0, PriceCacheWrite: 0,
121- BaseURL: "https://api.kimi.com/coding/v1",
120+ BaseURL: "https://api.kimi.com/coding",
122121 },
123122
124123 // ---- OpenAI / GPT-5 family ----
A internal/provider/openai_responses.go +65 -0
@@ -0,0 +1,65 @@
1+package provider
2+
3+// OpenAI Responses API client (api.openai.com/v1/responses).
4+//
5+// Strategy: reuse the existing codexClient (which already speaks the
6+// Responses wire format) via a header-rewriting RoundTripper. The codex
7+// client always sends `chatgpt-account-id` and `openai-beta`; the public
8+// Responses API on api.openai.com accepts but doesn't require those —
9+// we strip the account-id header on its way out so the request looks
10+// like a normal authenticated OpenAI call.
11+//
12+// This is a separate provider from `openai` (which is Chat Completions);
13+// users opt in by passing `--provider openai-responses` or by picking a
14+// model whose catalog entry tags it under that provider id.
15+//
16+// Auth: API key via Authorization: Bearer (set by codexClient itself).
17+// Different from openai-codex, which uses OAuth subscription tokens.
18+
19+import (
20+ "net/http"
21+ "strings"
22+)
23+
24+const openaiResponsesDefaultBaseURL = "https://api.openai.com/v1/responses"
25+
26+// openaiResponsesTransport rewrites headers so the codex client (which
27+// is designed for chatgpt.com OAuth tokens) can target the public
28+// OpenAI Responses API with a normal API key.
29+type openaiResponsesTransport struct {
30+ inner http.RoundTripper
31+}
32+
33+func (t *openaiResponsesTransport) RoundTrip(req *http.Request) (*http.Response, error) {
34+ clone := req.Clone(req.Context())
35+ // The codex client always sets chatgpt-account-id; api.openai.com
36+ // doesn't need it and may reject the OAuth identity headers, so
37+ // strip them.
38+ clone.Header.Del("chatgpt-account-id")
39+ clone.Header.Del("openai-beta")
40+ clone.Header.Del("originator")
41+ // Keep Authorization: Bearer <key> as set by the codex client.
42+ return t.inner.RoundTrip(clone)
43+}
44+
45+// NewOpenAIResponses returns an OpenAI Responses-API client (API-key
46+// flow). Uses the same wire format as the ChatGPT Codex backend but
47+// with the public api.openai.com endpoint and standard Bearer auth.
48+//
49+// baseURL may be empty; defaults to https://api.openai.com/v1/responses.
50+func NewOpenAIResponses(apiKey, baseURL string) Client {
51+ if baseURL == "" {
52+ baseURL = openaiResponsesDefaultBaseURL
53+ }
54+ httpClient := &http.Client{
55+ Transport: &openaiResponsesTransport{inner: http.DefaultTransport},
56+ Timeout: 0,
57+ }
58+ inner := &codexClient{
59+ token: apiKey,
60+ accountID: "", // unused; transport strips the header
61+ baseURL: strings.TrimRight(baseURL, "/"),
62+ http: httpClient,
63+ }
64+ return &renamedClient{inner: inner, name: "openai-responses"}
65+}
M internal/provider/provider_test.go +34 -0
@@ -78,6 +78,40 @@ func TestOpenAIErrorStatus(t *testing.T) {
7878 }
7979 }
8080
81+func TestAnthropicBuildRequestStripsAssistantImages(t *testing.T) {
82+ c := NewAnthropic("x", "").(*anthropicClient)
83+ wire, err := c.buildRequest(Request{
84+ Model: "claude-sonnet-4-5",
85+ Messages: []Message{
86+ {Role: RoleUser, Content: []Content{TextBlock{Text: "make an image"}}},
87+ {Role: RoleAssistant, Content: []Content{
88+ TextBlock{Text: "done"},
89+ ImageBlock{MimeType: "image/png", Data: []byte("png")},
90+ TextBlock{Text: "Saved image: `zot-gemini-image-x.png`"},
91+ }},
92+ {Role: RoleUser, Content: []Content{TextBlock{Text: "hello"}}},
93+ },
94+ })
95+ if err != nil {
96+ t.Fatal(err)
97+ }
98+ if len(wire.Messages) != 3 {
99+ t.Fatalf("messages=%d", len(wire.Messages))
100+ }
101+ assistant := wire.Messages[1]
102+ if assistant.Role != "assistant" {
103+ t.Fatalf("role=%q", assistant.Role)
104+ }
105+ if len(assistant.Content) != 2 {
106+ t.Fatalf("assistant content=%+v", assistant.Content)
107+ }
108+ for _, b := range assistant.Content {
109+ if _, ok := b.(anthImageBlock); ok {
110+ t.Fatalf("assistant image block was not stripped: %+v", assistant.Content)
111+ }
112+ }
113+}
114+
81115 func TestAnthropicStreamHappyPath(t *testing.T) {
82116 srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
83117 w.Header().Set("content-type", "text/event-stream")
M internal/provider/usermodels.go +35 -4
@@ -2,6 +2,7 @@ package provider
22
33 import (
44 "encoding/json"
5+ "fmt"
56 "os"
67 )
78
@@ -59,17 +60,34 @@ type UserModel struct {
5960 // (missing file, bad JSON, etc.) so the caller can treat it as
6061 // optional without error handling.
6162 func LoadUserModels(path string) []Model {
63+ models, _ := LoadUserModelsWithWarnings(path)
64+ return models
65+}
66+
67+// LoadUserModelsWithWarnings is like LoadUserModels but also returns
68+// human-readable warnings about every recoverable issue it found in
69+// the file (unknown provider id, empty model id, malformed JSON for a
70+// single provider block, etc.). The caller is responsible for
71+// surfacing the warnings; the file is never rejected wholesale unless
72+// the top-level JSON itself fails to parse.
73+func LoadUserModelsWithWarnings(path string) ([]Model, []string) {
74+ var warnings []string
6275 data, err := os.ReadFile(path)
6376 if err != nil {
64- return nil
77+ return nil, nil
6578 }
6679 var file UserModelsFile
6780 if err := json.Unmarshal(data, &file); err != nil {
68- return nil
81+ warnings = append(warnings, fmt.Sprintf("models.json: parse error: %v (file ignored)", err))
82+ return nil, warnings
6983 }
7084
7185 var out []Model
7286 for providerName, prov := range file.Providers {
87+ if providerName == "" {
88+ warnings = append(warnings, "models.json: empty provider key skipped")
89+ continue
90+ }
7391 // Normalize legacy transport aliases to their provider names.
7492 normalized := providerName
7593 switch providerName {
@@ -83,7 +101,20 @@ func LoadUserModels(path string) []Model {
83101 normalized = "deepseek"
84102 }
85103
86- for _, um := range prov.Models {
104+ for i, um := range prov.Models {
105+ if um.ID == "" {
106+ warnings = append(warnings, fmt.Sprintf("models.json: provider %q entry #%d has empty id; skipped", providerName, i))
107+ continue
108+ }
109+ if um.ContextWindow < 0 || um.MaxTokens < 0 {
110+ warnings = append(warnings, fmt.Sprintf("models.json: %s/%s has negative contextWindow/maxTokens; clamped to 0", normalized, um.ID))
111+ if um.ContextWindow < 0 {
112+ um.ContextWindow = 0
113+ }
114+ if um.MaxTokens < 0 {
115+ um.MaxTokens = 0
116+ }
117+ }
87118 m := Model{
88119 Provider: normalized,
89120 ID: um.ID,
@@ -104,7 +135,7 @@ func LoadUserModels(path string) []Model {
104135 out = append(out, m)
105136 }
106137 }
107- return out
138+ return out, warnings
108139 }
109140
110141 // SetUserModels merges user-defined models into the active catalog.
M internal/tui/view.go +21 -0
@@ -140,6 +140,15 @@ type View struct {
140140 // Streaming/in-flight work (v.Streaming, v.ToolCalls) is never
141141 // cached because it changes every delta.
142142 renderCache map[msgCacheKey][]string
143+
144+ // TailLimit caps how many messages from the END of Messages are
145+ // rendered. Messages older than the limit emit a zero-row slice
146+ // so first paint after a session resume doesn't pay the markdown
147+ // / chroma cost for the entire transcript at once. 0 means no
148+ // limit (render everything, the historical behaviour). Interactive
149+ // raises the cap when the user scrolls past the top of the
150+ // already-rendered tail.
151+ TailLimit int
143152 }
144153
145154 // msgCacheKey identifies a cached message render. hash is a 64-bit
@@ -280,7 +289,19 @@ func (v *View) BuildWithAnchors(width int) ([]string, []MessageAnchor) {
280289 // to visibly stutter while typing.
281290 rendered := make([][]string, len(v.Messages))
282291 total := 0
292+ // renderFrom is the first index we actually render. Anything
293+ // below it emits a zero-row placeholder; the message still gets
294+ // an anchor entry so /jump and scroll math keep working, and a
295+ // later Build with a higher TailLimit fills the gap.
296+ renderFrom := 0
297+ if v.TailLimit > 0 && len(v.Messages) > v.TailLimit {
298+ renderFrom = len(v.Messages) - v.TailLimit
299+ }
283300 for idx, m := range v.Messages {
301+ if idx < renderFrom {
302+ rendered[idx] = nil
303+ continue
304+ }
284305 // A turn is "open" once the assistant has started responding
285306 // to the most recent user prompt. Walk back over consecutive
286307 // assistant/tool messages: if any non-user message precedes

Keyboard shortcuts

Single-key shortcuts are on. ctrl k and esc work whatever this says.

ctrl k :
go to a section, branch, tag, file or repository
?
this list
j k
move the row cursor down / up
enter
open the row under the cursor; on Files, a directory opens or closes in place
↑ ↓ ← → home end
in the file tree (focus on a row): previous / next row, close / open a directory or go to its parent / first entry, first / last row. They never open a file
l h
on Files: open a directory or a file / close a directory or go up to its parent, then to the parent page. Elsewhere: into the side panel (on a merge request, its files) / back to the content; there j k move between its controls
g g G
first / last row
/
filter the list; on Files, find a file anywhere in the repository (esc clears, enter opens the best match)
b
switch branch or tag
g s g f g c g b g m g i g r g a
tabs: summary, files, commits, branches & tags, merge requests, issues, releases, actions (when available)
g h
home
w
wide mode: use the whole width of the screen, or not (remembered)
y
copy a permalink pinned to the commit
d u x
diff: split / unified / collapse or expand all
e E
diff: the whole file in view / every file, with the lines the hunks hide
n
merge request: next note in the diff
esc
close a dialog, leave a field

Branches 1

Tags 476

↑↓ moveenter openesc close