1package provider
2
3import (
4 "fmt"
5 "maps"
6 "sync"
7)
8
9// Model describes a single LLM we know about.
10type Model struct {
11 Provider string // "anthropic" | "openai"
12 ID string // API id
13 DisplayName string
14 API string // wire API override for providers that support multiple protocols
15 ContextWindow int
16 MaxOutput int
17 Reasoning bool // supports reasoning
18 ReasoningLevelMap map[string]string // optional level overrides; empty values remove a level
19
20 // AdaptiveThinking marks Anthropic models that only support the
21 // adaptive thinking mode (Opus 4.7+). These reject explicit
22 // thinking budgets (thinking:{type:"enabled",budget_tokens:N} -> 400)
23 // and also reject non-default sampling params (temperature/top_p/
24 // top_k). The Anthropic client sends thinking:{type:"adaptive"} plus
25 // output_config.effort and omits temperature for these models.
26 AdaptiveThinking bool
27
28 // AdaptiveThinkingCompat marks Anthropic-compatible models that require
29 // thinking:{type:"adaptive"} but still accept sampling parameters and do
30 // not support Anthropic's output_config.effort extension.
31 AdaptiveThinkingCompat bool
32
33 // Prices are USD per 1M tokens. Price*Above fields apply to all token
34 // classes when the total prompt size exceeds PriceTierInputTokens.
35 PriceInput float64
36 PriceOutput float64
37 PriceCacheRead float64
38 PriceCacheWrite float64
39 PriceTierInputTokens int
40 PriceInputAbove float64
41 PriceOutputAbove float64
42 PriceCacheReadAbove float64
43 PriceCacheWriteAbove float64
44
45 // Speculative marks models whose ids are known from the upstream
46 // vendor's CLI but not yet live on their public API. They'll 404
47 // today but start working the moment the provider flips the switch.
48 Speculative bool
49
50 // BaseURL overrides the provider's default API endpoint for this
51 // model. Optional; when empty the provider's default (or the
52 // --base-url flag) is used. Useful for local models served by
53 // ollama, vLLM, LM Studio, etc.
54 BaseURL string
55
56 // Source is where this model entry came from: "catalog" (baked in),
57 // "live" (discovered via /v1/models), or "cache" (loaded from the
58 // on-disk cache). Informational.
59 Source string
60}
61
62// Catalog is the hardcoded, read-only list of supported models.
63// Prices are USD per 1M tokens. The list is curated to what zot's
64// clients (Anthropic Messages + OpenAI Chat Completions) can actually
65// talk to; models that are only reachable through the OpenAI Responses
66// API (o1-pro, o3-pro, gpt-5-pro) are omitted.
67var Catalog = []Model{
68 // ---- Anthropic / Claude 4.x ----
69 {
70 Provider: "anthropic", ID: "claude-sonnet-4-5", DisplayName: "Claude Sonnet 4.5 (latest)",
71 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
72 PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, PriceCacheWrite: 3.75,
73 },
74 {
75 Provider: "anthropic", ID: "claude-opus-4-1", DisplayName: "Claude Opus 4.1 (latest)",
76 ContextWindow: 200000, MaxOutput: 32000, Reasoning: true,
77 PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, PriceCacheWrite: 18.75,
78 },
79 {
80 Provider: "anthropic", ID: "claude-opus-4-0", DisplayName: "Claude Opus 4 (latest)",
81 ContextWindow: 200000, MaxOutput: 32000, Reasoning: true,
82 PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, PriceCacheWrite: 18.75,
83 },
84 {
85 Provider: "anthropic", ID: "claude-sonnet-4-0", DisplayName: "Claude Sonnet 4 (latest)",
86 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
87 PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, PriceCacheWrite: 3.75,
88 },
89 {
90 Provider: "anthropic", ID: "claude-haiku-4-5", DisplayName: "Claude Haiku 4.5 (latest)",
91 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
92 PriceInput: 1, PriceOutput: 5, PriceCacheRead: 0.1, PriceCacheWrite: 1.25,
93 },
94
95 // ---- Anthropic / Claude 3.x (legacy) ----
96 {
97 Provider: "anthropic", ID: "claude-3-7-sonnet-20250219", DisplayName: "Claude Sonnet 3.7",
98 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
99 PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, PriceCacheWrite: 3.75,
100 },
101 {
102 Provider: "anthropic", ID: "claude-3-5-sonnet-20241022", DisplayName: "Claude Sonnet 3.5 v2",
103 ContextWindow: 200000, MaxOutput: 8192, Reasoning: false,
104 PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, PriceCacheWrite: 3.75,
105 },
106 {
107 Provider: "anthropic", ID: "claude-3-5-haiku-latest", DisplayName: "Claude Haiku 3.5 (latest)",
108 ContextWindow: 200000, MaxOutput: 8192, Reasoning: false,
109 PriceInput: 0.8, PriceOutput: 4, PriceCacheRead: 0.08, PriceCacheWrite: 1,
110 },
111 {
112 Provider: "anthropic", ID: "claude-3-opus-20240229", DisplayName: "Claude Opus 3",
113 ContextWindow: 200000, MaxOutput: 4096, Reasoning: false,
114 PriceInput: 15, PriceOutput: 75, PriceCacheRead: 1.5, PriceCacheWrite: 18.75,
115 },
116
117 // ---- DeepSeek ----
118 // The current public DeepSeek API exposes the V4 family on
119 // api.deepseek.com/v1. Pro is the flagship reasoning model;
120 // Flash is the cheaper/faster sibling. Both accept image inputs
121 // (multimodal parts: image_url) in addition to text.
122 {
123 Provider: "deepseek", ID: "deepseek-v4-pro", DisplayName: "DeepSeek V4 Pro",
124 ContextWindow: 1000000, MaxOutput: 384000, Reasoning: true,
125 PriceInput: 0.435, PriceOutput: 0.87, PriceCacheRead: 0.003625,
126 BaseURL: "https://api.deepseek.com",
127 },
128 {
129 Provider: "deepseek", ID: "deepseek-v4-flash", DisplayName: "DeepSeek V4 Flash",
130 ContextWindow: 1000000, MaxOutput: 384000, Reasoning: true,
131 PriceInput: 0.14, PriceOutput: 0.28, PriceCacheRead: 0.0028,
132 BaseURL: "https://api.deepseek.com",
133 },
134
135 // ---- Kimi / Kimi Code ----
136 // Anthropic-messages on https://api.kimi.com/coding (no /v1 suffix;
137 // the Anthropic client appends /v1/messages itself).
138 {
139 Provider: "kimi", ID: "kimi-for-coding", DisplayName: "Kimi For Coding",
140 ContextWindow: 262144, MaxOutput: 32768, Reasoning: true,
141 PriceInput: 0, PriceOutput: 0, PriceCacheRead: 0,
142 BaseURL: "https://api.kimi.com/coding",
143 },
144
145 // ---- OpenAI / GPT-5 family ----
146 {
147 Provider: "openai", ID: "gpt-5", DisplayName: "GPT-5",
148 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
149 PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125,
150 },
151 {
152 Provider: "openai", ID: "gpt-5-mini", DisplayName: "GPT-5 Mini",
153 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
154 PriceInput: 0.25, PriceOutput: 2, PriceCacheRead: 0.025,
155 },
156 {
157 Provider: "openai", ID: "gpt-5-nano", DisplayName: "GPT-5 Nano",
158 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
159 PriceInput: 0.05, PriceOutput: 0.4, PriceCacheRead: 0.005,
160 },
161
162 // ---- OpenAI / GPT-4.1 family ----
163 {
164 Provider: "openai", ID: "gpt-4.1", DisplayName: "GPT-4.1",
165 ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false,
166 PriceInput: 2, PriceOutput: 8, PriceCacheRead: 0.5,
167 },
168 {
169 Provider: "openai", ID: "gpt-4.1-mini", DisplayName: "GPT-4.1 mini",
170 ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false,
171 PriceInput: 0.4, PriceOutput: 1.6, PriceCacheRead: 0.1,
172 },
173 {
174 Provider: "openai", ID: "gpt-4.1-nano", DisplayName: "GPT-4.1 nano",
175 ContextWindow: 1047576, MaxOutput: 32768, Reasoning: false,
176 PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.03,
177 },
178
179 // ---- OpenAI / GPT-4o family ----
180 {
181 Provider: "openai", ID: "gpt-4o", DisplayName: "GPT-4o",
182 ContextWindow: 128000, MaxOutput: 16384, Reasoning: false,
183 PriceInput: 2.5, PriceOutput: 10, PriceCacheRead: 1.25,
184 },
185 {
186 Provider: "openai", ID: "gpt-4o-mini", DisplayName: "GPT-4o mini",
187 ContextWindow: 128000, MaxOutput: 16384, Reasoning: false,
188 PriceInput: 0.15, PriceOutput: 0.6, PriceCacheRead: 0.08,
189 },
190
191 // ---- OpenAI / reasoning models ----
192 {
193 Provider: "openai", ID: "o4-mini", DisplayName: "o4-mini",
194 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
195 PriceInput: 1.1, PriceOutput: 4.4, PriceCacheRead: 0.28,
196 },
197 {
198 Provider: "openai", ID: "o3", DisplayName: "o3",
199 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
200 PriceInput: 2, PriceOutput: 8, PriceCacheRead: 0.5,
201 },
202 {
203 Provider: "openai", ID: "o3-mini", DisplayName: "o3-mini",
204 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
205 PriceInput: 1.1, PriceOutput: 4.4, PriceCacheRead: 0.55,
206 },
207 {
208 Provider: "openai", ID: "o1", DisplayName: "o1",
209 ContextWindow: 200000, MaxOutput: 100000, Reasoning: true,
210 PriceInput: 15, PriceOutput: 60, PriceCacheRead: 7.5,
211 },
212
213 // ---- Google / Gemini ----
214 {
215 Provider: "google", ID: "gemini-2.5-pro", DisplayName: "Gemini 2.5 Pro",
216 ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true,
217 PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.125,
218 },
219 {
220 Provider: "google", ID: "gemini-2.5-flash", DisplayName: "Gemini 2.5 Flash",
221 ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true,
222 PriceInput: 0.3, PriceOutput: 2.5, PriceCacheRead: 0.03,
223 },
224 {
225 Provider: "google", ID: "gemini-2.5-flash-lite", DisplayName: "Gemini 2.5 Flash-Lite",
226 ContextWindow: 1048576, MaxOutput: 65536, Reasoning: true,
227 PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.01,
228 },
229 {
230 Provider: "google", ID: "gemini-2.0-flash", DisplayName: "Gemini 2.0 Flash",
231 ContextWindow: 1048576, MaxOutput: 8192, Reasoning: false,
232 PriceInput: 0.1, PriceOutput: 0.4, PriceCacheRead: 0.025,
233 },
234 {
235 Provider: "google", ID: "gemini-2.0-flash-lite", DisplayName: "Gemini 2.0 Flash-Lite",
236 ContextWindow: 1048576, MaxOutput: 8192, Reasoning: false,
237 PriceInput: 0.075, PriceOutput: 0.3, PriceCacheRead: 0,
238 },
239
240 // ---- OpenRouter ----
241 // Seed entry only: the default model returned by defaultModelForProvider
242 // so it resolves offline. The full OpenRouter catalog is discovered live
243 // (DiscoverOpenRouter) and overlays this on refresh.
244 {
245 Provider: "openrouter", ID: "anthropic/claude-sonnet-4.5", DisplayName: "Claude Sonnet 4.5 (OpenRouter)",
246 ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true,
247 PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, PriceCacheWrite: 3.75,
248 BaseURL: openrouterDefaultBaseURL,
249 },
250 {
251 Provider: "openrouter", ID: "anthropic/claude-sonnet-5", DisplayName: "Claude Sonnet 5 (OpenRouter)",
252 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
253 PriceInput: 2, PriceOutput: 10, PriceCacheRead: 0.2, PriceCacheWrite: 2.5,
254 BaseURL: openrouterDefaultBaseURL,
255 },
256
257 // ---- Gondola ----
258 // Seed entries only: kimi-k3 is the default returned by
259 // defaultModelForProvider so it resolves offline. Gondola is a USDC-paid
260 // marketplace gateway for Venice AI inference; prices/limits pulled live
261 // from https://api.gondola-ai.com/v1/models (2026-08-11).
262 {
263 Provider: "gondola", ID: "kimi-k3", DisplayName: "Kimi K3 (Gondola)",
264 ContextWindow: 1000000, MaxOutput: 131072, Reasoning: true,
265 PriceInput: 1.1475, PriceOutput: 5.7375, PriceCacheRead: 0.11475,
266 BaseURL: gondolaDefaultBaseURL,
267 },
268 {
269 Provider: "gondola", ID: "claude-opus-5", DisplayName: "Claude Opus 5 (Gondola)",
270 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
271 PriceInput: 1.836, PriceOutput: 9.18, PriceCacheRead: 0.1836, PriceCacheWrite: 2.295,
272 BaseURL: gondolaDefaultBaseURL,
273 },
274
275 // ---- Anthropic / newer families ----
276 {
277 Provider: "anthropic", ID: "claude-opus-4-5", DisplayName: "Claude Opus 4.5 (latest)",
278 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
279 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
280 Speculative: true,
281 },
282 {
283 Provider: "anthropic", ID: "claude-opus-4-6", DisplayName: "Claude Opus 4.6",
284 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true,
285 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
286 Speculative: true,
287 },
288 {
289 Provider: "anthropic", ID: "claude-opus-4-7", DisplayName: "Claude Opus 4.7",
290 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
291 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
292 Speculative: true,
293 },
294 {
295 Provider: "anthropic", ID: "claude-opus-4-8", DisplayName: "Claude Opus 4.8",
296 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
297 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
298 Speculative: true,
299 },
300 {
301 Provider: "anthropic", ID: "claude-opus-5", DisplayName: "Claude Opus 5",
302 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
303 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
304 },
305 {
306 Provider: "anthropic", ID: "claude-sonnet-4-6", DisplayName: "Claude Sonnet 4.6",
307 ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true,
308 PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, PriceCacheWrite: 3.75,
309 Speculative: true,
310 },
311 {
312 Provider: "anthropic", ID: "claude-sonnet-5", DisplayName: "Claude Sonnet 5",
313 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
314 PriceInput: 2, PriceOutput: 10, PriceCacheRead: 0.2, PriceCacheWrite: 2.5,
315 },
316 {
317 Provider: "anthropic", ID: "claude-fable-5", DisplayName: "Claude Fable 5",
318 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
319 PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5,
320 },
321 {
322 Provider: "anthropic", ID: "claude-fable-5-1", DisplayName: "Claude Fable 5.1",
323 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
324 PriceInput: 10, PriceOutput: 50, PriceCacheRead: 0.25, PriceCacheWrite: 12.5,
325 },
326
327 // ---- Speculative: OpenAI ----
328 // Public OpenAI API route. The ChatGPT/Codex subscription route is
329 // represented separately below as provider "openai-codex".
330 {
331 Provider: "openai", ID: "gpt-5.1", DisplayName: "GPT-5.1",
332 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
333 PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.13,
334 Speculative: true,
335 },
336 {
337 Provider: "openai", ID: "gpt-5.2", DisplayName: "GPT-5.2",
338 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
339 PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175,
340 Speculative: true,
341 },
342 {
343 Provider: "openai", ID: "gpt-5.4", DisplayName: "GPT-5.4",
344 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
345 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
346 Speculative: true,
347 },
348 {
349 Provider: "openai", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
350 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
351 PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075,
352 Speculative: true,
353 },
354 {
355 Provider: "openai", ID: "gpt-5.5", DisplayName: "GPT-5.5",
356 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
357 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
358 Speculative: true,
359 },
360 {
361 Provider: "openai", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses,
362 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
363 PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1, PriceCacheWrite: 1.25,
364 Speculative: true,
365 },
366 {
367 Provider: "openai", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol", API: APIResponses,
368 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
369 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
370 Speculative: true,
371 },
372 {
373 Provider: "openai", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra", API: APIResponses,
374 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
375 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25, PriceCacheWrite: 3.125,
376 Speculative: true,
377 },
378
379 // ---- OpenAI / GPT-6 ----
380 {
381 Provider: "openai", ID: "gpt-6-astra", DisplayName: "GPT-6 Astra", API: APIResponses,
382 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
383 PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5,
384 PriceTierInputTokens: 272000,
385 PriceInputAbove: 20, PriceOutputAbove: 75, PriceCacheReadAbove: 2, PriceCacheWriteAbove: 25,
386 },
387
388 // ---- OpenAI Codex / ChatGPT subscription backend ----
389 // Same model ids as the OpenAI family, but routed through the
390 // ChatGPT Codex OAuth backend rather than api.openai.com.
391 {
392 Provider: "openai-codex", ID: "gpt-5.3-codex-spark", DisplayName: "GPT-5.3 Codex Spark",
393 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
394 PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175,
395 Speculative: true,
396 },
397 {
398 Provider: "openai-codex", ID: "gpt-5.4", DisplayName: "GPT-5.4",
399 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
400 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
401 Speculative: true,
402 },
403 {
404 Provider: "openai-codex", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
405 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
406 PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075,
407 Speculative: true,
408 },
409 {
410 Provider: "openai-codex", ID: "gpt-5.5", DisplayName: "GPT-5.5",
411 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
412 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
413 Speculative: true,
414 },
415 {
416 Provider: "openai-codex", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna",
417 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
418 PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1, PriceCacheWrite: 1.25,
419 Speculative: true,
420 },
421 {
422 Provider: "openai-codex", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol",
423 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
424 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
425 Speculative: true,
426 },
427 {
428 Provider: "openai-codex", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra",
429 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
430 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25, PriceCacheWrite: 3.125,
431 Speculative: true,
432 },
433 {
434 Provider: "openai-codex", ID: "gpt-6-astra", DisplayName: "GPT-6 Astra", API: APIResponses,
435 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
436 PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5,
437 PriceTierInputTokens: 272000,
438 PriceInputAbove: 20, PriceOutputAbove: 75, PriceCacheReadAbove: 2, PriceCacheWriteAbove: 25,
439 },
440}
441
442// DefaultModel is used when the user does not specify one.
443var DefaultModel = Catalog[0] // claude-sonnet-4-5
444
445// ----- active (merged) catalog -----
446//
447// Callers should use Active() / FindModel / ModelsForProvider for
448// lookups. They return the baked-in Catalog merged with any live
449// models loaded via SetLiveModels.
450
451var (
452 activeMu sync.RWMutex
453 active []Model // live overlay merged in via SetLiveModels; nil = none yet
454 activeSet bool // true once SetLiveModels has run (even with empty live)
455 managedModels []Model // ephemeral models exposed by local model managers
456)
457
458// SetLiveModels replaces the "live" overlay used by the active catalog.
459// Typically called after a successful /v1/models discovery or on load
460// from the on-disk cache.
461func SetLiveModels(live []Model) {
462 activeMu.Lock()
463 defer activeMu.Unlock()
464 activeSet = true
465 if len(live) == 0 {
466 active = nil
467 return
468 }
469 active = MergeCatalog(live)
470}
471
472// Active returns the current merged catalog.
473//
474// When no live overlay has been set it returns the fully-assembled
475// static Catalog. Reading Catalog here (rather than capturing it into
476// a package-level var initializer) is load-bearing: the extended
477// catalog in catalog_builtin.go / extra_models.go is appended from
478// init() functions, which run AFTER package-level var initializers.
479// Snapshotting Catalog at var-init time would freeze the picker to the
480// curated seed list and drop every extra provider (openrouter, groq,
481// xai, ...). Deferring the read to call time avoids that ordering trap.
482func Active() []Model {
483 activeMu.RLock()
484 defer activeMu.RUnlock()
485 src := active
486 if !activeSet || src == nil {
487 src = Catalog
488 }
489 out := make([]Model, len(src))
490 copy(out, src)
491 for i := range out {
492 out[i].ReasoningLevelMap = maps.Clone(out[i].ReasoningLevelMap)
493 }
494 if len(managedModels) == 0 {
495 return out
496 }
497 index := make(map[string]int, len(out))
498 for i, model := range out {
499 index[model.Provider+"\x00"+model.ID] = i
500 }
501 for _, model := range managedModels {
502 model.ReasoningLevelMap = maps.Clone(model.ReasoningLevelMap)
503 key := model.Provider + "\x00" + model.ID
504 if i, ok := index[key]; ok {
505 out[i] = model
506 continue
507 }
508 index[key] = len(out)
509 out = append(out, model)
510 }
511 return out
512}
513
514// SetManagedModels replaces the ephemeral catalog entries supplied by local
515// model managers. Unlike SetLiveModels, this does not disturb provider model
516// discovery or its on-disk cache.
517func SetManagedModels(models []Model) {
518 activeMu.Lock()
519 defer activeMu.Unlock()
520 managedModels = append([]Model(nil), models...)
521}
522
523// FindModel returns a Model by id, optionally constrained by provider.
524// If provider is empty, the first matching id is returned. Looks up
525// against the merged active catalog.
526func FindModel(provider, id string) (Model, error) {
527 for _, m := range Active() {
528 if m.ID == id && (provider == "" || m.Provider == provider) {
529 return m, nil
530 }
531 }
532 return Model{}, fmt.Errorf("unknown model %q (provider=%q)", id, provider)
533}
534
535// ModelsForProvider returns all models for the given provider, from the
536// merged active catalog.
537func ModelsForProvider(provider string) []Model {
538 var out []Model
539 for _, m := range Active() {
540 if m.Provider == provider {
541 out = append(out, m)
542 }
543 }
544 return out
545}
546
547// ComputeCost returns the USD cost for the given usage on model m.
548func ComputeCost(m Model, u Usage) float64 {
549 inputPrice := m.PriceInput
550 outputPrice := m.PriceOutput
551 cacheReadPrice := m.PriceCacheRead
552 cacheWritePrice := m.PriceCacheWrite
553 promptTokens := u.InputTokens + u.CacheReadTokens + u.CacheWriteTokens
554 if m.PriceTierInputTokens > 0 && promptTokens > m.PriceTierInputTokens {
555 inputPrice = m.PriceInputAbove
556 outputPrice = m.PriceOutputAbove
557 cacheReadPrice = m.PriceCacheReadAbove
558 cacheWritePrice = m.PriceCacheWriteAbove
559 }
560
561 const per = 1_000_000.0
562 return float64(u.InputTokens)*inputPrice/per +
563 float64(u.OutputTokens)*outputPrice/per +
564 float64(u.CacheReadTokens)*cacheReadPrice/per +
565 float64(u.CacheWriteTokens)*cacheWritePrice/per
566}