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 // ---- Yolo-Auto ----
276 // Seed entries only: `qwen3.8-flash` is the free and paid default returned by
277 // defaultModelForProvider so it resolves offline. Yolo-Auto is an
278 // OpenAI-compatible gateway on flat-rate plans with no per-token
279 // billing, so no per-token prices are listed. The limits are the
280 // provider's documented conservative client settings; the full catalog,
281 // including each key's plan-bounded window, comes from /v1/models
282 // (DiscoverYoloAuto).
283 {
284 Provider: "yolo-auto", ID: "qwen3.8-flash", DisplayName: "Qwen3.8 Flash",
285 ContextWindow: yoloAutoContextWindow, MaxOutput: yoloAutoMaxOutput,
286 Reasoning: true,
287 BaseURL: yoloAutoDefaultBaseURL,
288 },
289 {
290 Provider: "yolo-auto", ID: "yolo", DisplayName: "Yolo",
291 ContextWindow: yoloAutoContextWindow, MaxOutput: yoloAutoMaxOutput,
292 Reasoning: true,
293 BaseURL: yoloAutoDefaultBaseURL,
294 },
295 {
296 Provider: "yolo-auto", ID: "yolo-small", DisplayName: "Yolo Small",
297 ContextWindow: yoloAutoContextWindow, MaxOutput: yoloAutoMaxOutput,
298 BaseURL: yoloAutoDefaultBaseURL,
299 },
300
301 // ---- Anthropic / newer families ----
302 {
303 Provider: "anthropic", ID: "claude-opus-4-5", DisplayName: "Claude Opus 4.5 (latest)",
304 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
305 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
306 Speculative: true,
307 },
308 {
309 Provider: "anthropic", ID: "claude-opus-4-6", DisplayName: "Claude Opus 4.6",
310 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true,
311 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
312 Speculative: true,
313 },
314 {
315 Provider: "anthropic", ID: "claude-opus-4-7", DisplayName: "Claude Opus 4.7",
316 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
317 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
318 Speculative: true,
319 },
320 {
321 Provider: "anthropic", ID: "claude-opus-4-8", DisplayName: "Claude Opus 4.8",
322 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
323 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
324 Speculative: true,
325 },
326 {
327 Provider: "anthropic", ID: "claude-opus-5", DisplayName: "Claude Opus 5",
328 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
329 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
330 },
331 {
332 Provider: "anthropic", ID: "claude-opus-5-5", DisplayName: "Claude Opus 5.5",
333 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
334 PriceInput: 4, PriceOutput: 20, PriceCacheRead: 0.2, PriceCacheWrite: 5,
335 },
336 {
337 Provider: "anthropic", ID: "claude-sonnet-4-6", DisplayName: "Claude Sonnet 4.6",
338 ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true,
339 PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, PriceCacheWrite: 3.75,
340 Speculative: true,
341 },
342 {
343 Provider: "anthropic", ID: "claude-sonnet-5", DisplayName: "Claude Sonnet 5",
344 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
345 PriceInput: 2, PriceOutput: 10, PriceCacheRead: 0.2, PriceCacheWrite: 2.5,
346 },
347 {
348 Provider: "anthropic", ID: "claude-fable-5", DisplayName: "Claude Fable 5",
349 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
350 PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5,
351 },
352 {
353 Provider: "anthropic", ID: "claude-fable-5-1", DisplayName: "Claude Fable 5.1",
354 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
355 PriceInput: 10, PriceOutput: 50, PriceCacheRead: 0.25, PriceCacheWrite: 12.5,
356 },
357
358 // ---- Speculative: OpenAI ----
359 // Public OpenAI API route. The ChatGPT/Codex subscription route is
360 // represented separately below as provider "openai-codex".
361 {
362 Provider: "openai", ID: "gpt-5.1", DisplayName: "GPT-5.1",
363 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
364 PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.13,
365 Speculative: true,
366 },
367 {
368 Provider: "openai", ID: "gpt-5.2", DisplayName: "GPT-5.2",
369 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
370 PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175,
371 Speculative: true,
372 },
373 {
374 Provider: "openai", ID: "gpt-5.4", DisplayName: "GPT-5.4",
375 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
376 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
377 Speculative: true,
378 },
379 {
380 Provider: "openai", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
381 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
382 PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075,
383 Speculative: true,
384 },
385 {
386 Provider: "openai", ID: "gpt-5.5", DisplayName: "GPT-5.5",
387 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
388 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
389 Speculative: true,
390 },
391 {
392 Provider: "openai", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses,
393 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
394 PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1, PriceCacheWrite: 1.25,
395 Speculative: true,
396 },
397 {
398 Provider: "openai", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol", API: APIResponses,
399 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
400 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
401 Speculative: true,
402 },
403 {
404 Provider: "openai", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra", API: APIResponses,
405 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
406 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25, PriceCacheWrite: 3.125,
407 Speculative: true,
408 },
409
410 // ---- OpenAI / GPT-6 ----
411 {
412 Provider: "openai", ID: "gpt-6-astra", DisplayName: "GPT-6 Astra", API: APIResponses,
413 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
414 PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5,
415 PriceTierInputTokens: 272000,
416 PriceInputAbove: 20, PriceOutputAbove: 75, PriceCacheReadAbove: 2, PriceCacheWriteAbove: 25,
417 },
418
419 // ---- OpenAI Codex / ChatGPT subscription backend ----
420 // Same model ids as the OpenAI family, but routed through the
421 // ChatGPT Codex OAuth backend rather than api.openai.com.
422 {
423 Provider: "openai-codex", ID: "gpt-5.3-codex-spark", DisplayName: "GPT-5.3 Codex Spark",
424 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
425 PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175,
426 Speculative: true,
427 },
428 {
429 Provider: "openai-codex", ID: "gpt-5.4", DisplayName: "GPT-5.4",
430 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
431 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
432 Speculative: true,
433 },
434 {
435 Provider: "openai-codex", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
436 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
437 PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075,
438 Speculative: true,
439 },
440 {
441 Provider: "openai-codex", ID: "gpt-5.5", DisplayName: "GPT-5.5",
442 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
443 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
444 Speculative: true,
445 },
446 {
447 Provider: "openai-codex", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna",
448 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
449 PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1, PriceCacheWrite: 1.25,
450 Speculative: true,
451 },
452 {
453 Provider: "openai-codex", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol",
454 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
455 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
456 Speculative: true,
457 },
458 {
459 Provider: "openai-codex", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra",
460 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
461 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25, PriceCacheWrite: 3.125,
462 Speculative: true,
463 },
464 {
465 Provider: "openai-codex", ID: "gpt-6-astra", DisplayName: "GPT-6 Astra", API: APIResponses,
466 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
467 PriceInput: 10, PriceOutput: 50, PriceCacheRead: 1, PriceCacheWrite: 12.5,
468 PriceTierInputTokens: 272000,
469 PriceInputAbove: 20, PriceOutputAbove: 75, PriceCacheReadAbove: 2, PriceCacheWriteAbove: 25,
470 },
471}
472
473// DefaultModel is used when the user does not specify one.
474var DefaultModel = Catalog[0] // claude-sonnet-4-5
475
476// ----- active (merged) catalog -----
477//
478// Callers should use Active() / FindModel / ModelsForProvider for
479// lookups. They return the baked-in Catalog merged with any live
480// models loaded via SetLiveModels.
481
482var (
483 activeMu sync.RWMutex
484 active []Model // live overlay merged in via SetLiveModels; nil = none yet
485 activeSet bool // true once SetLiveModels has run (even with empty live)
486 managedModels []Model // ephemeral models exposed by local model managers
487 modelAvailability = map[string]map[string]bool{} // account-specific, memory only
488)
489
490// SetLiveModels replaces the "live" overlay used by the active catalog.
491// Typically called after a successful /v1/models discovery or on load
492// from the on-disk cache.
493func SetLiveModels(live []Model) {
494 activeMu.Lock()
495 defer activeMu.Unlock()
496 activeSet = true
497 if len(live) == 0 {
498 active = nil
499 return
500 }
501 active = MergeCatalog(live)
502}
503
504// Active returns the current merged catalog.
505//
506// When no live overlay has been set it returns the fully-assembled
507// static Catalog. Reading Catalog here (rather than capturing it into
508// a package-level var initializer) is load-bearing: the extended
509// catalog in catalog_builtin.go / extra_models.go is appended from
510// init() functions, which run AFTER package-level var initializers.
511// Snapshotting Catalog at var-init time would freeze the picker to the
512// curated seed list and drop every extra provider (openrouter, groq,
513// xai, ...). Deferring the read to call time avoids that ordering trap.
514func Active() []Model { return activeModels(true) }
515
516func activeModels(availableOnly bool) []Model {
517 activeMu.RLock()
518 defer activeMu.RUnlock()
519 src := active
520 if !activeSet || src == nil {
521 src = Catalog
522 }
523 out := make([]Model, len(src))
524 copy(out, src)
525 for i := range out {
526 out[i].ReasoningLevelMap = maps.Clone(out[i].ReasoningLevelMap)
527 }
528 if len(managedModels) == 0 {
529 if availableOnly {
530 return filterAvailableModels(out)
531 }
532 return out
533 }
534 index := make(map[string]int, len(out))
535 for i, model := range out {
536 index[model.Provider+"\x00"+model.ID] = i
537 }
538 for _, model := range managedModels {
539 model.ReasoningLevelMap = maps.Clone(model.ReasoningLevelMap)
540 key := model.Provider + "\x00" + model.ID
541 if i, ok := index[key]; ok {
542 if out[i].Source == "user" {
543 continue
544 }
545 out[i] = model
546 continue
547 }
548 index[key] = len(out)
549 out = append(out, model)
550 }
551 if availableOnly {
552 return filterAvailableModels(out)
553 }
554 return out
555}
556
557// SetModelAvailability restricts a provider's visible catalog to the supplied
558// account model IDs. A nil slice clears the restriction; an empty non-nil slice
559// hides all models. This snapshot is not persisted in the shared model cache.
560func SetModelAvailability(provider string, ids []string) {
561 activeMu.Lock()
562 defer activeMu.Unlock()
563 if ids == nil {
564 delete(modelAvailability, provider)
565 return
566 }
567 allowed := make(map[string]bool, len(ids))
568 for _, id := range ids {
569 allowed[id] = true
570 }
571 modelAvailability[provider] = allowed
572}
573
574// Caller holds activeMu. Models is already a private copy.
575func filterAvailableModels(models []Model) []Model {
576 out := models[:0]
577 for _, m := range models {
578 allowed, restricted := modelAvailability[m.Provider]
579 if !restricted || allowed[m.ID] {
580 out = append(out, m)
581 }
582 }
583 return out
584}
585
586// SetManagedModels replaces the ephemeral catalog entries supplied by local
587// model managers. Unlike SetLiveModels, this does not disturb provider model
588// discovery or its on-disk cache.
589func SetManagedModels(models []Model) {
590 activeMu.Lock()
591 defer activeMu.Unlock()
592 managedModels = append([]Model(nil), models...)
593}
594
595// SetManagedModelsForProvider replaces only one local server's entries.
596func SetManagedModelsForProvider(providerID string, models []Model) {
597 activeMu.Lock()
598 defer activeMu.Unlock()
599 kept := managedModels[:0]
600 for _, m := range managedModels {
601 if m.Provider != providerID {
602 kept = append(kept, m)
603 }
604 }
605 managedModels = kept
606 for _, m := range models {
607 if m.Provider == providerID {
608 managedModels = append(managedModels, m)
609 }
610 }
611}
612
613// FindModel returns a Model by id, optionally constrained by provider.
614// If provider is empty, the first matching id is returned. Looks up
615// against the merged catalog, including models hidden by account availability.
616// Existing sessions still need stable wire and pricing metadata after a refresh.
617func FindModel(provider, id string) (Model, error) {
618 for _, m := range activeModels(false) {
619 if m.ID == id && (provider == "" || m.Provider == provider) {
620 return m, nil
621 }
622 }
623 return Model{}, fmt.Errorf("unknown model %q (provider=%q)", id, provider)
624}
625
626// ModelsForProvider returns all models for the given provider, from the
627// merged active catalog.
628func ModelsForProvider(provider string) []Model {
629 var out []Model
630 for _, m := range Active() {
631 if m.Provider == provider {
632 out = append(out, m)
633 }
634 }
635 return out
636}
637
638// ComputeCost returns the USD cost for the given usage on model m.
639func ComputeCost(m Model, u Usage) float64 {
640 inputPrice := m.PriceInput
641 outputPrice := m.PriceOutput
642 cacheReadPrice := m.PriceCacheRead
643 cacheWritePrice := m.PriceCacheWrite
644 promptTokens := u.InputTokens + u.CacheReadTokens + u.CacheWriteTokens
645 if m.PriceTierInputTokens > 0 && promptTokens > m.PriceTierInputTokens {
646 inputPrice = m.PriceInputAbove
647 outputPrice = m.PriceOutputAbove
648 cacheReadPrice = m.PriceCacheReadAbove
649 cacheWritePrice = m.PriceCacheWriteAbove
650 }
651
652 const per = 1_000_000.0
653 return float64(u.InputTokens)*inputPrice/per +
654 float64(u.OutputTokens)*outputPrice/per +
655 float64(u.CacheReadTokens)*cacheReadPrice/per +
656 float64(u.CacheWriteTokens)*cacheWritePrice/per
657}