1package provider
2
3import (
4 "fmt"
5 "sync"
6)
7
8// Model describes a single LLM we know about.
9type Model struct {
10 Provider string // "anthropic" | "openai"
11 ID string // API id
12 DisplayName string
13 ContextWindow int
14 MaxOutput int
15 Reasoning bool // supports reasoning/thinking
16
17 // Prices are USD per 1M tokens.
18 PriceInput float64
19 PriceOutput float64
20 PriceCacheRead float64
21 PriceCacheWrite float64
22
23 // Speculative marks models whose ids are known from the upstream
24 // vendor's CLI but not yet live on their public API. They'll 404
25 // today but start working the moment the provider flips the switch.
26 Speculative bool
27
28 // BaseURL overrides the provider's default API endpoint for this
29 // model. Optional; when empty the provider's default (or the
30 // --base-url flag) is used. Useful for local models served by
31 // ollama, vLLM, LM Studio, etc.
32 BaseURL string
33
34 // Source is where this model entry came from: "catalog" (baked in),
35 // "live" (discovered via /v1/models), or "cache" (loaded from the
36 // on-disk cache). Informational.
37 Source string
38}
39
40// Catalog is the hardcoded, read-only list of supported models.
41// Prices are USD per 1M tokens. The list is curated to what zot's
42// clients (Anthropic Messages + OpenAI Chat Completions) can actually
43// talk to; models that are only reachable through the OpenAI Responses
44// API (o1-pro, o3-pro, gpt-5-pro) are omitted.
45var Catalog = []Model{
46 // ---- Anthropic / Claude 4.x ----
47 {
48 Provider: "anthropic", ID: "claude-sonnet-4-5", DisplayName: "Claude Sonnet 4.5",
49 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
50 PriceInput: 3.00, PriceOutput: 15.00, PriceCacheRead: 0.30, PriceCacheWrite: 3.75,
51 },
52 {
53 Provider: "anthropic", ID: "claude-opus-4-1", DisplayName: "Claude Opus 4.1",
54 ContextWindow: 200000, MaxOutput: 32000, Reasoning: true,
55 PriceInput: 15.00, PriceOutput: 75.00, PriceCacheRead: 1.50, PriceCacheWrite: 18.75,
56 },
57 {
58 Provider: "anthropic", ID: "claude-opus-4-0", DisplayName: "Claude Opus 4",
59 ContextWindow: 200000, MaxOutput: 32000, Reasoning: true,
60 PriceInput: 15.00, PriceOutput: 75.00, PriceCacheRead: 1.50, PriceCacheWrite: 18.75,
61 },
62 {
63 Provider: "anthropic", ID: "claude-sonnet-4-0", DisplayName: "Claude Sonnet 4",
64 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
65 PriceInput: 3.00, PriceOutput: 15.00, PriceCacheRead: 0.30, PriceCacheWrite: 3.75,
66 },
67 {
68 Provider: "anthropic", ID: "claude-haiku-4-5", DisplayName: "Claude Haiku 4.5",
69 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
70 PriceInput: 1.00, PriceOutput: 5.00, PriceCacheRead: 0.10, PriceCacheWrite: 1.25,
71 },
72
73 // ---- Anthropic / Claude 3.x (legacy) ----
74 {
75 Provider: "anthropic", ID: "claude-3-7-sonnet-20250219", DisplayName: "Claude Sonnet 3.7",
76 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
77 PriceInput: 3.00, PriceOutput: 15.00, PriceCacheRead: 0.30, PriceCacheWrite: 3.75,
78 },
79 {
80 Provider: "anthropic", ID: "claude-3-5-sonnet-20241022", DisplayName: "Claude Sonnet 3.5 v2",
81 ContextWindow: 200000, MaxOutput: 8192, Reasoning: false,
82 PriceInput: 3.00, PriceOutput: 15.00, PriceCacheRead: 0.30, PriceCacheWrite: 3.75,
83 },
84 {
85 Provider: "anthropic", ID: "claude-3-5-haiku-latest", DisplayName: "Claude Haiku 3.5",
86 ContextWindow: 200000, MaxOutput: 8192, Reasoning: false,
87 PriceInput: 0.80, PriceOutput: 4.00, PriceCacheRead: 0.08, PriceCacheWrite: 1.00,
88 },
89 {
90 Provider: "anthropic", ID: "claude-3-opus-20240229", DisplayName: "Claude Opus 3",
91 ContextWindow: 200000, MaxOutput: 4096, Reasoning: false,
92 PriceInput: 15.00, PriceOutput: 75.00, PriceCacheRead: 1.50, PriceCacheWrite: 18.75,
93 },
94
95 // ---- Kimi / Kimi Code ----
96 {
97 Provider: "kimi", ID: "kimi-for-coding", DisplayName: "Kimi-k2.6",
98 ContextWindow: 262144, MaxOutput: 32000, Reasoning: true,
99 BaseURL: "https://api.kimi.com/coding/v1",
100 },
101
102 // ---- OpenAI / GPT-5 family ----
103 {
104 Provider: "openai", ID: "gpt-5", DisplayName: "GPT-5",
105 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
106 PriceInput: 1.25, PriceOutput: 10.00, PriceCacheRead: 0.125,
107 },
108 {
109 Provider: "openai", ID: "gpt-5-mini", DisplayName: "GPT-5 mini",
110 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
111 PriceInput: 0.25, PriceOutput: 2.00, PriceCacheRead: 0.025,
112 },
113 {
114 Provider: "openai", ID: "gpt-5-nano", DisplayName: "GPT-5 nano",
115 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
116 PriceInput: 0.05, PriceOutput: 0.40, PriceCacheRead: 0.005,
117 },
118
119 // ---- OpenAI / GPT-4.1 family ----
120 {
121 Provider: "openai", ID: "gpt-4.1", DisplayName: "GPT-4.1",
122 ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false,
123 PriceInput: 2.00, PriceOutput: 8.00, PriceCacheRead: 0.50,
124 },
125 {
126 Provider: "openai", ID: "gpt-4.1-mini", DisplayName: "GPT-4.1 mini",
127 ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false,
128 PriceInput: 0.40, PriceOutput: 1.60, PriceCacheRead: 0.10,
129 },
130 {
131 Provider: "openai", ID: "gpt-4.1-nano", DisplayName: "GPT-4.1 nano",
132 ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false,
133 PriceInput: 0.10, PriceOutput: 0.40, PriceCacheRead: 0.03,
134 },
135
136 // ---- OpenAI / GPT-4o family ----
137 {
138 Provider: "openai", ID: "gpt-4o", DisplayName: "GPT-4o",
139 ContextWindow: 128000, MaxOutput: 16384, Reasoning: false,
140 PriceInput: 2.50, PriceOutput: 10.00, PriceCacheRead: 1.25,
141 },
142 {
143 Provider: "openai", ID: "gpt-4o-mini", DisplayName: "GPT-4o mini",
144 ContextWindow: 128000, MaxOutput: 16384, Reasoning: false,
145 PriceInput: 0.15, PriceOutput: 0.60, PriceCacheRead: 0.08,
146 },
147
148 // ---- OpenAI / reasoning models ----
149 {
150 Provider: "openai", ID: "o4-mini", DisplayName: "o4-mini",
151 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
152 PriceInput: 1.10, PriceOutput: 4.40, PriceCacheRead: 0.275,
153 },
154 {
155 Provider: "openai", ID: "o3", DisplayName: "o3",
156 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
157 PriceInput: 2.00, PriceOutput: 8.00, PriceCacheRead: 0.50,
158 },
159 {
160 Provider: "openai", ID: "o3-mini", DisplayName: "o3-mini",
161 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
162 PriceInput: 1.10, PriceOutput: 4.40, PriceCacheRead: 0.55,
163 },
164 {
165 Provider: "openai", ID: "o1", DisplayName: "o1",
166 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
167 PriceInput: 15.00, PriceOutput: 60.00, PriceCacheRead: 7.50,
168 },
169
170 // ---- Speculative: Anthropic ----
171 {
172 Provider: "anthropic", ID: "claude-opus-4-5", DisplayName: "Claude Opus 4.5",
173 // 200k ctx / 64k maxOutput per Anthropic's published sizing
174 // for the opus-4-5 family; the 1M context is a 4.6+ thing.
175 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
176 PriceInput: 5.00, PriceOutput: 25.00, PriceCacheRead: 0.50, PriceCacheWrite: 6.25,
177 Speculative: true,
178 },
179 {
180 Provider: "anthropic", ID: "claude-opus-4-6", DisplayName: "Claude Opus 4.6",
181 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true,
182 PriceInput: 5.00, PriceOutput: 25.00, PriceCacheRead: 0.50, PriceCacheWrite: 6.25,
183 Speculative: true,
184 },
185 {
186 Provider: "anthropic", ID: "claude-opus-4-7", DisplayName: "Claude Opus 4.7",
187 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true,
188 PriceInput: 5.00, PriceOutput: 25.00, PriceCacheRead: 0.50, PriceCacheWrite: 6.25,
189 Speculative: true,
190 },
191 {
192 Provider: "anthropic", ID: "claude-sonnet-4-6", DisplayName: "Claude Sonnet 4.6",
193 ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true,
194 PriceInput: 3.00, PriceOutput: 15.00, PriceCacheRead: 0.30, PriceCacheWrite: 3.75,
195 Speculative: true,
196 },
197
198 // ---- Speculative: OpenAI ----
199 // Context windows on the OpenAI gpt-5.x family differ by route:
200 // the direct API advertises 400k, the ChatGPT Codex OAuth backend
201 // caps at 272k. zot serves both auth modes from one catalog row
202 // per id, so we pin to the smaller number to keep the context-usage
203 // meter honest under subscription auth. Users on the direct API
204 // simply see a conservative headroom estimate.
205 {
206 Provider: "openai", ID: "gpt-5.1", DisplayName: "GPT-5.1",
207 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
208 PriceInput: 1.25, PriceOutput: 10.00, PriceCacheRead: 0.125,
209 Speculative: true,
210 },
211 {
212 Provider: "openai", ID: "gpt-5.2", DisplayName: "GPT-5.2",
213 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
214 PriceInput: 1.75, PriceOutput: 14.00, PriceCacheRead: 0.175,
215 Speculative: true,
216 },
217 {
218 Provider: "openai", ID: "gpt-5.3", DisplayName: "GPT-5.3",
219 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
220 PriceInput: 1.75, PriceOutput: 14.00, PriceCacheRead: 0.175,
221 Speculative: true,
222 },
223 {
224 Provider: "openai", ID: "gpt-5.4", DisplayName: "GPT-5.4",
225 // ContextWindow: 272k across every route we support (OpenAI
226 // direct API and the ChatGPT Codex OAuth backend).
227 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
228 PriceInput: 2.50, PriceOutput: 15.00, PriceCacheRead: 0.25,
229 Speculative: true,
230 },
231 {
232 Provider: "openai", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
233 // ContextWindow: 400k on the OpenAI direct API, 272k on the
234 // ChatGPT Codex OAuth backend. We pin to the smaller Codex
235 // cap so the context-usage meter is honest under subscription
236 // auth; direct-API users simply see a conservative headroom
237 // estimate rather than an inflated one.
238 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
239 PriceInput: 0.75, PriceOutput: 4.50, PriceCacheRead: 0.075,
240 Speculative: true,
241 },
242 {
243 Provider: "openai", ID: "gpt-5.5", DisplayName: "GPT-5.5",
244 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
245 PriceInput: 2.50, PriceOutput: 15.00, PriceCacheRead: 0.25,
246 Speculative: true,
247 },
248 {
249 Provider: "openai", ID: "gpt-5.5-mini", DisplayName: "GPT-5.5 mini",
250 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
251 PriceInput: 0.75, PriceOutput: 4.50, PriceCacheRead: 0.075,
252 Speculative: true,
253 },
254}
255
256// DefaultModel is used when the user does not specify one.
257var DefaultModel = Catalog[0] // claude-sonnet-4-5
258
259// ----- active (merged) catalog -----
260//
261// Callers should use Active() / FindModel / ModelsForProvider for
262// lookups. They return the baked-in Catalog merged with any live
263// models loaded via SetLiveModels.
264
265var (
266 activeMu sync.RWMutex
267 active []Model = Catalog // default: just the static catalog
268)
269
270// SetLiveModels replaces the "live" overlay used by the active catalog.
271// Typically called after a successful /v1/models discovery or on load
272// from the on-disk cache.
273func SetLiveModels(live []Model) {
274 activeMu.Lock()
275 defer activeMu.Unlock()
276 if len(live) == 0 {
277 active = Catalog
278 return
279 }
280 active = MergeCatalog(live)
281}
282
283// Active returns the current merged catalog.
284func Active() []Model {
285 activeMu.RLock()
286 defer activeMu.RUnlock()
287 out := make([]Model, len(active))
288 copy(out, active)
289 return out
290}
291
292// FindModel returns a Model by id, optionally constrained by provider.
293// If provider is empty, the first matching id is returned. Looks up
294// against the merged active catalog.
295func FindModel(provider, id string) (Model, error) {
296 for _, m := range Active() {
297 if m.ID == id && (provider == "" || m.Provider == provider) {
298 return m, nil
299 }
300 }
301 return Model{}, fmt.Errorf("unknown model %q (provider=%q)", id, provider)
302}
303
304// ModelsForProvider returns all models for the given provider, from the
305// merged active catalog.
306func ModelsForProvider(provider string) []Model {
307 var out []Model
308 for _, m := range Active() {
309 if m.Provider == provider {
310 out = append(out, m)
311 }
312 }
313 return out
314}
315
316// ComputeCost returns the USD cost for the given usage on model m.
317func ComputeCost(m Model, u Usage) float64 {
318 const per = 1_000_000.0
319 return float64(u.InputTokens)*m.PriceInput/per +
320 float64(u.OutputTokens)*m.PriceOutput/per +
321 float64(u.CacheReadTokens)*m.PriceCacheRead/per +
322 float64(u.CacheWriteTokens)*m.PriceCacheWrite/per
323}