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 // ---- OpenAI / GPT-5 family ----
96 {
97 Provider: "openai", ID: "gpt-5", DisplayName: "GPT-5",
98 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
99 PriceInput: 1.25, PriceOutput: 10.00, PriceCacheRead: 0.125,
100 },
101 {
102 Provider: "openai", ID: "gpt-5-mini", DisplayName: "GPT-5 mini",
103 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
104 PriceInput: 0.25, PriceOutput: 2.00, PriceCacheRead: 0.025,
105 },
106 {
107 Provider: "openai", ID: "gpt-5-nano", DisplayName: "GPT-5 nano",
108 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
109 PriceInput: 0.05, PriceOutput: 0.40, PriceCacheRead: 0.005,
110 },
111
112 // ---- OpenAI / GPT-4.1 family ----
113 {
114 Provider: "openai", ID: "gpt-4.1", DisplayName: "GPT-4.1",
115 ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false,
116 PriceInput: 2.00, PriceOutput: 8.00, PriceCacheRead: 0.50,
117 },
118 {
119 Provider: "openai", ID: "gpt-4.1-mini", DisplayName: "GPT-4.1 mini",
120 ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false,
121 PriceInput: 0.40, PriceOutput: 1.60, PriceCacheRead: 0.10,
122 },
123 {
124 Provider: "openai", ID: "gpt-4.1-nano", DisplayName: "GPT-4.1 nano",
125 ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false,
126 PriceInput: 0.10, PriceOutput: 0.40, PriceCacheRead: 0.03,
127 },
128
129 // ---- OpenAI / GPT-4o family ----
130 {
131 Provider: "openai", ID: "gpt-4o", DisplayName: "GPT-4o",
132 ContextWindow: 128000, MaxOutput: 16384, Reasoning: false,
133 PriceInput: 2.50, PriceOutput: 10.00, PriceCacheRead: 1.25,
134 },
135 {
136 Provider: "openai", ID: "gpt-4o-mini", DisplayName: "GPT-4o mini",
137 ContextWindow: 128000, MaxOutput: 16384, Reasoning: false,
138 PriceInput: 0.15, PriceOutput: 0.60, PriceCacheRead: 0.08,
139 },
140
141 // ---- OpenAI / reasoning models ----
142 {
143 Provider: "openai", ID: "o4-mini", DisplayName: "o4-mini",
144 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
145 PriceInput: 1.10, PriceOutput: 4.40, PriceCacheRead: 0.275,
146 },
147 {
148 Provider: "openai", ID: "o3", DisplayName: "o3",
149 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
150 PriceInput: 2.00, PriceOutput: 8.00, PriceCacheRead: 0.50,
151 },
152 {
153 Provider: "openai", ID: "o3-mini", DisplayName: "o3-mini",
154 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
155 PriceInput: 1.10, PriceOutput: 4.40, PriceCacheRead: 0.55,
156 },
157 {
158 Provider: "openai", ID: "o1", DisplayName: "o1",
159 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
160 PriceInput: 15.00, PriceOutput: 60.00, PriceCacheRead: 7.50,
161 },
162
163 // ---- Speculative: Anthropic ----
164 {
165 Provider: "anthropic", ID: "claude-opus-4-5", DisplayName: "Claude Opus 4.5",
166 // 200k ctx / 64k maxOutput per Anthropic's published sizing
167 // for the opus-4-5 family; the 1M context is a 4.6+ thing.
168 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
169 PriceInput: 5.00, PriceOutput: 25.00, PriceCacheRead: 0.50, PriceCacheWrite: 6.25,
170 Speculative: true,
171 },
172 {
173 Provider: "anthropic", ID: "claude-opus-4-6", DisplayName: "Claude Opus 4.6",
174 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true,
175 PriceInput: 5.00, PriceOutput: 25.00, PriceCacheRead: 0.50, PriceCacheWrite: 6.25,
176 Speculative: true,
177 },
178 {
179 Provider: "anthropic", ID: "claude-opus-4-7", DisplayName: "Claude Opus 4.7",
180 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true,
181 PriceInput: 5.00, PriceOutput: 25.00, PriceCacheRead: 0.50, PriceCacheWrite: 6.25,
182 Speculative: true,
183 },
184 {
185 Provider: "anthropic", ID: "claude-sonnet-4-6", DisplayName: "Claude Sonnet 4.6",
186 ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true,
187 PriceInput: 3.00, PriceOutput: 15.00, PriceCacheRead: 0.30, PriceCacheWrite: 3.75,
188 Speculative: true,
189 },
190
191 // ---- Speculative: OpenAI ----
192 // Context windows on the OpenAI gpt-5.x family differ by route:
193 // the direct API advertises 400k, the ChatGPT Codex OAuth backend
194 // caps at 272k. zot serves both auth modes from one catalog row
195 // per id, so we pin to the smaller number to keep the context-usage
196 // meter honest under subscription auth. Users on the direct API
197 // simply see a conservative headroom estimate.
198 {
199 Provider: "openai", ID: "gpt-5.1", DisplayName: "GPT-5.1",
200 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
201 PriceInput: 1.25, PriceOutput: 10.00, PriceCacheRead: 0.125,
202 Speculative: true,
203 },
204 {
205 Provider: "openai", ID: "gpt-5.2", DisplayName: "GPT-5.2",
206 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
207 PriceInput: 1.75, PriceOutput: 14.00, PriceCacheRead: 0.175,
208 Speculative: true,
209 },
210 {
211 Provider: "openai", ID: "gpt-5.3", DisplayName: "GPT-5.3",
212 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
213 PriceInput: 1.75, PriceOutput: 14.00, PriceCacheRead: 0.175,
214 Speculative: true,
215 },
216 {
217 Provider: "openai", ID: "gpt-5.4", DisplayName: "GPT-5.4",
218 // ContextWindow: 272k across every route we support (OpenAI
219 // direct API and the ChatGPT Codex OAuth backend).
220 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
221 PriceInput: 2.50, PriceOutput: 15.00, PriceCacheRead: 0.25,
222 Speculative: true,
223 },
224 {
225 Provider: "openai", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
226 // ContextWindow: 400k on the OpenAI direct API, 272k on the
227 // ChatGPT Codex OAuth backend. We pin to the smaller Codex
228 // cap so the context-usage meter is honest under subscription
229 // auth; direct-API users simply see a conservative headroom
230 // estimate rather than an inflated one.
231 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
232 PriceInput: 0.75, PriceOutput: 4.50, PriceCacheRead: 0.075,
233 Speculative: true,
234 },
235 {
236 Provider: "openai", ID: "gpt-5.5", DisplayName: "GPT-5.5",
237 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
238 PriceInput: 2.50, PriceOutput: 15.00, PriceCacheRead: 0.25,
239 Speculative: true,
240 },
241 {
242 Provider: "openai", ID: "gpt-5.5-mini", DisplayName: "GPT-5.5 mini",
243 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
244 PriceInput: 0.75, PriceOutput: 4.50, PriceCacheRead: 0.075,
245 Speculative: true,
246 },
247}
248
249// DefaultModel is used when the user does not specify one.
250var DefaultModel = Catalog[0] // claude-sonnet-4-5
251
252// ----- active (merged) catalog -----
253//
254// Callers should use Active() / FindModel / ModelsForProvider for
255// lookups. They return the baked-in Catalog merged with any live
256// models loaded via SetLiveModels.
257
258var (
259 activeMu sync.RWMutex
260 active []Model = Catalog // default: just the static catalog
261)
262
263// SetLiveModels replaces the "live" overlay used by the active catalog.
264// Typically called after a successful /v1/models discovery or on load
265// from the on-disk cache.
266func SetLiveModels(live []Model) {
267 activeMu.Lock()
268 defer activeMu.Unlock()
269 if len(live) == 0 {
270 active = Catalog
271 return
272 }
273 active = MergeCatalog(live)
274}
275
276// Active returns the current merged catalog.
277func Active() []Model {
278 activeMu.RLock()
279 defer activeMu.RUnlock()
280 out := make([]Model, len(active))
281 copy(out, active)
282 return out
283}
284
285// FindModel returns a Model by id, optionally constrained by provider.
286// If provider is empty, the first matching id is returned. Looks up
287// against the merged active catalog.
288func FindModel(provider, id string) (Model, error) {
289 for _, m := range Active() {
290 if m.ID == id && (provider == "" || m.Provider == provider) {
291 return m, nil
292 }
293 }
294 return Model{}, fmt.Errorf("unknown model %q (provider=%q)", id, provider)
295}
296
297// ModelsForProvider returns all models for the given provider, from the
298// merged active catalog.
299func ModelsForProvider(provider string) []Model {
300 var out []Model
301 for _, m := range Active() {
302 if m.Provider == provider {
303 out = append(out, m)
304 }
305 }
306 return out
307}
308
309// ComputeCost returns the USD cost for the given usage on model m.
310func ComputeCost(m Model, u Usage) float64 {
311 const per = 1_000_000.0
312 return float64(u.InputTokens)*m.PriceInput/per +
313 float64(u.OutputTokens)*m.PriceOutput/per +
314 float64(u.CacheReadTokens)*m.PriceCacheRead/per +
315 float64(u.CacheWriteTokens)*m.PriceCacheWrite/per
316}