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 // ---- Speculative: Anthropic ----
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: 0.5, PriceCacheWrite: 6.25,
320 Speculative: true,
321 },
322
323 // ---- Speculative: OpenAI ----
324 // Public OpenAI API route. The ChatGPT/Codex subscription route is
325 // represented separately below as provider "openai-codex".
326 {
327 Provider: "openai", ID: "gpt-5.1", DisplayName: "GPT-5.1",
328 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
329 PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.13,
330 Speculative: true,
331 },
332 {
333 Provider: "openai", ID: "gpt-5.2", DisplayName: "GPT-5.2",
334 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
335 PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175,
336 Speculative: true,
337 },
338 {
339 Provider: "openai", ID: "gpt-5.4", DisplayName: "GPT-5.4",
340 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
341 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
342 Speculative: true,
343 },
344 {
345 Provider: "openai", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
346 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
347 PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075,
348 Speculative: true,
349 },
350 {
351 Provider: "openai", ID: "gpt-5.5", DisplayName: "GPT-5.5",
352 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
353 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
354 Speculative: true,
355 },
356 {
357 Provider: "openai", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses,
358 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
359 PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1, PriceCacheWrite: 1.25,
360 Speculative: true,
361 },
362 {
363 Provider: "openai", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol", API: APIResponses,
364 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
365 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
366 Speculative: true,
367 },
368 {
369 Provider: "openai", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra", API: APIResponses,
370 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
371 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25, PriceCacheWrite: 3.125,
372 Speculative: true,
373 },
374
375 // ---- OpenAI Codex / ChatGPT subscription backend ----
376 // Same model ids as the OpenAI family, but routed through the
377 // ChatGPT Codex OAuth backend rather than api.openai.com.
378 {
379 Provider: "openai-codex", ID: "gpt-5.3-codex-spark", DisplayName: "GPT-5.3 Codex Spark",
380 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
381 PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175,
382 Speculative: true,
383 },
384 {
385 Provider: "openai-codex", ID: "gpt-5.4", DisplayName: "GPT-5.4",
386 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
387 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
388 Speculative: true,
389 },
390 {
391 Provider: "openai-codex", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
392 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
393 PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075,
394 Speculative: true,
395 },
396 {
397 Provider: "openai-codex", ID: "gpt-5.5", DisplayName: "GPT-5.5",
398 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
399 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
400 Speculative: true,
401 },
402 {
403 Provider: "openai-codex", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna",
404 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
405 PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1, PriceCacheWrite: 1.25,
406 Speculative: true,
407 },
408 {
409 Provider: "openai-codex", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol",
410 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
411 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
412 Speculative: true,
413 },
414 {
415 Provider: "openai-codex", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra",
416 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
417 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25, PriceCacheWrite: 3.125,
418 Speculative: true,
419 },
420}
421
422// DefaultModel is used when the user does not specify one.
423var DefaultModel = Catalog[0] // claude-sonnet-4-5
424
425// ----- active (merged) catalog -----
426//
427// Callers should use Active() / FindModel / ModelsForProvider for
428// lookups. They return the baked-in Catalog merged with any live
429// models loaded via SetLiveModels.
430
431var (
432 activeMu sync.RWMutex
433 active []Model // live overlay merged in via SetLiveModels; nil = none yet
434 activeSet bool // true once SetLiveModels has run (even with empty live)
435 managedModels []Model // ephemeral models exposed by local model managers
436)
437
438// SetLiveModels replaces the "live" overlay used by the active catalog.
439// Typically called after a successful /v1/models discovery or on load
440// from the on-disk cache.
441func SetLiveModels(live []Model) {
442 activeMu.Lock()
443 defer activeMu.Unlock()
444 activeSet = true
445 if len(live) == 0 {
446 active = nil
447 return
448 }
449 active = MergeCatalog(live)
450}
451
452// Active returns the current merged catalog.
453//
454// When no live overlay has been set it returns the fully-assembled
455// static Catalog. Reading Catalog here (rather than capturing it into
456// a package-level var initializer) is load-bearing: the extended
457// catalog in catalog_builtin.go / extra_models.go is appended from
458// init() functions, which run AFTER package-level var initializers.
459// Snapshotting Catalog at var-init time would freeze the picker to the
460// curated seed list and drop every extra provider (openrouter, groq,
461// xai, ...). Deferring the read to call time avoids that ordering trap.
462func Active() []Model {
463 activeMu.RLock()
464 defer activeMu.RUnlock()
465 src := active
466 if !activeSet || src == nil {
467 src = Catalog
468 }
469 out := make([]Model, len(src))
470 copy(out, src)
471 for i := range out {
472 out[i].ReasoningLevelMap = maps.Clone(out[i].ReasoningLevelMap)
473 }
474 if len(managedModels) == 0 {
475 return out
476 }
477 index := make(map[string]int, len(out))
478 for i, model := range out {
479 index[model.Provider+"\x00"+model.ID] = i
480 }
481 for _, model := range managedModels {
482 model.ReasoningLevelMap = maps.Clone(model.ReasoningLevelMap)
483 key := model.Provider + "\x00" + model.ID
484 if i, ok := index[key]; ok {
485 out[i] = model
486 continue
487 }
488 index[key] = len(out)
489 out = append(out, model)
490 }
491 return out
492}
493
494// SetManagedModels replaces the ephemeral catalog entries supplied by local
495// model managers. Unlike SetLiveModels, this does not disturb provider model
496// discovery or its on-disk cache.
497func SetManagedModels(models []Model) {
498 activeMu.Lock()
499 defer activeMu.Unlock()
500 managedModels = append([]Model(nil), models...)
501}
502
503// FindModel returns a Model by id, optionally constrained by provider.
504// If provider is empty, the first matching id is returned. Looks up
505// against the merged active catalog.
506func FindModel(provider, id string) (Model, error) {
507 for _, m := range Active() {
508 if m.ID == id && (provider == "" || m.Provider == provider) {
509 return m, nil
510 }
511 }
512 return Model{}, fmt.Errorf("unknown model %q (provider=%q)", id, provider)
513}
514
515// ModelsForProvider returns all models for the given provider, from the
516// merged active catalog.
517func ModelsForProvider(provider string) []Model {
518 var out []Model
519 for _, m := range Active() {
520 if m.Provider == provider {
521 out = append(out, m)
522 }
523 }
524 return out
525}
526
527// ComputeCost returns the USD cost for the given usage on model m.
528func ComputeCost(m Model, u Usage) float64 {
529 inputPrice := m.PriceInput
530 outputPrice := m.PriceOutput
531 cacheReadPrice := m.PriceCacheRead
532 cacheWritePrice := m.PriceCacheWrite
533 promptTokens := u.InputTokens + u.CacheReadTokens + u.CacheWriteTokens
534 if m.PriceTierInputTokens > 0 && promptTokens > m.PriceTierInputTokens {
535 inputPrice = m.PriceInputAbove
536 outputPrice = m.PriceOutputAbove
537 cacheReadPrice = m.PriceCacheReadAbove
538 cacheWritePrice = m.PriceCacheWriteAbove
539 }
540
541 const per = 1_000_000.0
542 return float64(u.InputTokens)*inputPrice/per +
543 float64(u.OutputTokens)*outputPrice/per +
544 float64(u.CacheReadTokens)*cacheReadPrice/per +
545 float64(u.CacheWriteTokens)*cacheWritePrice/per
546}