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 zut'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 // ---- Speculative: Anthropic ----
258 {
259 Provider: "anthropic", ID: "claude-opus-4-5", DisplayName: "Claude Opus 4.5 (latest)",
260 ContextWindow: 200000, MaxOutput: 64000, Reasoning: true,
261 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
262 Speculative: true,
263 },
264 {
265 Provider: "anthropic", ID: "claude-opus-4-6", DisplayName: "Claude Opus 4.6",
266 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true,
267 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
268 Speculative: true,
269 },
270 {
271 Provider: "anthropic", ID: "claude-opus-4-7", DisplayName: "Claude Opus 4.7",
272 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
273 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
274 Speculative: true,
275 },
276 {
277 Provider: "anthropic", ID: "claude-opus-4-8", DisplayName: "Claude Opus 4.8",
278 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
279 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
280 Speculative: true,
281 },
282 {
283 Provider: "anthropic", ID: "claude-opus-5", DisplayName: "Claude Opus 5",
284 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
285 PriceInput: 5, PriceOutput: 25, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
286 },
287 {
288 Provider: "anthropic", ID: "claude-sonnet-4-6", DisplayName: "Claude Sonnet 4.6",
289 ContextWindow: 1000000, MaxOutput: 64000, Reasoning: true,
290 PriceInput: 3, PriceOutput: 15, PriceCacheRead: 0.3, PriceCacheWrite: 3.75,
291 Speculative: true,
292 },
293 {
294 Provider: "anthropic", ID: "claude-sonnet-5", DisplayName: "Claude Sonnet 5",
295 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
296 PriceInput: 2, PriceOutput: 10, PriceCacheRead: 0.2, PriceCacheWrite: 2.5,
297 },
298 {
299 Provider: "anthropic", ID: "claude-fable-5", DisplayName: "Claude Fable 5",
300 ContextWindow: 1000000, MaxOutput: 128000, Reasoning: true, AdaptiveThinking: true,
301 PriceInput: 10, PriceOutput: 50, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
302 Speculative: true,
303 },
304
305 // ---- Speculative: OpenAI ----
306 // Public OpenAI API route. The ChatGPT/Codex subscription route is
307 // represented separately below as provider "openai-codex".
308 {
309 Provider: "openai", ID: "gpt-5.1", DisplayName: "GPT-5.1",
310 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
311 PriceInput: 1.25, PriceOutput: 10, PriceCacheRead: 0.13,
312 Speculative: true,
313 },
314 {
315 Provider: "openai", ID: "gpt-5.2", DisplayName: "GPT-5.2",
316 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
317 PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175,
318 Speculative: true,
319 },
320 {
321 Provider: "openai", ID: "gpt-5.4", DisplayName: "GPT-5.4",
322 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
323 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
324 Speculative: true,
325 },
326 {
327 Provider: "openai", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
328 ContextWindow: 400000, MaxOutput: 128000, Reasoning: true,
329 PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075,
330 Speculative: true,
331 },
332 {
333 Provider: "openai", ID: "gpt-5.5", DisplayName: "GPT-5.5",
334 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
335 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
336 Speculative: true,
337 },
338 {
339 Provider: "openai", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses,
340 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
341 PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1, PriceCacheWrite: 1.25,
342 Speculative: true,
343 },
344 {
345 Provider: "openai", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol", API: APIResponses,
346 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
347 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
348 Speculative: true,
349 },
350 {
351 Provider: "openai", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra", API: APIResponses,
352 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
353 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25, PriceCacheWrite: 3.125,
354 Speculative: true,
355 },
356
357 // ---- OpenAI Codex / ChatGPT subscription backend ----
358 // Same model ids as the OpenAI family, but routed through the
359 // ChatGPT Codex OAuth backend rather than api.openai.com.
360 {
361 Provider: "openai-codex", ID: "gpt-5.3-codex-spark", DisplayName: "GPT-5.3 Codex Spark",
362 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
363 PriceInput: 1.75, PriceOutput: 14, PriceCacheRead: 0.175,
364 Speculative: true,
365 },
366 {
367 Provider: "openai-codex", ID: "gpt-5.4", DisplayName: "GPT-5.4",
368 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
369 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25,
370 Speculative: true,
371 },
372 {
373 Provider: "openai-codex", ID: "gpt-5.4-mini", DisplayName: "GPT-5.4 mini",
374 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
375 PriceInput: 0.75, PriceOutput: 4.5, PriceCacheRead: 0.075,
376 Speculative: true,
377 },
378 {
379 Provider: "openai-codex", ID: "gpt-5.5", DisplayName: "GPT-5.5",
380 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
381 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5,
382 Speculative: true,
383 },
384 {
385 Provider: "openai-codex", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna",
386 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
387 PriceInput: 1, PriceOutput: 6, PriceCacheRead: 0.1, PriceCacheWrite: 1.25,
388 Speculative: true,
389 },
390 {
391 Provider: "openai-codex", ID: "gpt-5.6-sol", DisplayName: "GPT-5.6 Sol",
392 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
393 PriceInput: 5, PriceOutput: 30, PriceCacheRead: 0.5, PriceCacheWrite: 6.25,
394 Speculative: true,
395 },
396 {
397 Provider: "openai-codex", ID: "gpt-5.6-terra", DisplayName: "GPT-5.6 Terra",
398 ContextWindow: 272000, MaxOutput: 128000, Reasoning: true,
399 PriceInput: 2.5, PriceOutput: 15, PriceCacheRead: 0.25, PriceCacheWrite: 3.125,
400 Speculative: true,
401 },
402}
403
404// DefaultModel is used when the user does not specify one.
405var DefaultModel = Catalog[0] // claude-sonnet-4-5
406
407// ----- active (merged) catalog -----
408//
409// Callers should use Active() / FindModel / ModelsForProvider for
410// lookups. They return the baked-in Catalog merged with any live
411// models loaded via SetLiveModels.
412
413var (
414 activeMu sync.RWMutex
415 active []Model // live overlay merged in via SetLiveModels; nil = none yet
416 activeSet bool // true once SetLiveModels has run (even with empty live)
417 managedModels []Model // ephemeral models exposed by local model managers
418)
419
420// SetLiveModels replaces the "live" overlay used by the active catalog.
421// Typically called after a successful /v1/models discovery or on load
422// from the on-disk cache.
423func SetLiveModels(live []Model) {
424 activeMu.Lock()
425 defer activeMu.Unlock()
426 activeSet = true
427 if len(live) == 0 {
428 active = nil
429 return
430 }
431 active = MergeCatalog(live)
432}
433
434// Active returns the current merged catalog.
435//
436// When no live overlay has been set it returns the fully-assembled
437// static Catalog. Reading Catalog here (rather than capturing it into
438// a package-level var initializer) is load-bearing: the extended
439// catalog in catalog_builtin.go / extra_models.go is appended from
440// init() functions, which run AFTER package-level var initializers.
441// Snapshotting Catalog at var-init time would freeze the picker to the
442// curated seed list and drop every extra provider (openrouter, groq,
443// xai, ...). Deferring the read to call time avoids that ordering trap.
444func Active() []Model {
445 activeMu.RLock()
446 defer activeMu.RUnlock()
447 src := active
448 if !activeSet || src == nil {
449 src = Catalog
450 }
451 out := make([]Model, len(src))
452 copy(out, src)
453 for i := range out {
454 out[i].ReasoningLevelMap = maps.Clone(out[i].ReasoningLevelMap)
455 }
456 if len(managedModels) == 0 {
457 return out
458 }
459 index := make(map[string]int, len(out))
460 for i, model := range out {
461 index[model.Provider+"\x00"+model.ID] = i
462 }
463 for _, model := range managedModels {
464 model.ReasoningLevelMap = maps.Clone(model.ReasoningLevelMap)
465 key := model.Provider + "\x00" + model.ID
466 if i, ok := index[key]; ok {
467 out[i] = model
468 continue
469 }
470 index[key] = len(out)
471 out = append(out, model)
472 }
473 return out
474}
475
476// SetManagedModels replaces the ephemeral catalog entries supplied by local
477// model managers. Unlike SetLiveModels, this does not disturb provider model
478// discovery or its on-disk cache.
479func SetManagedModels(models []Model) {
480 activeMu.Lock()
481 defer activeMu.Unlock()
482 managedModels = append([]Model(nil), models...)
483}
484
485// FindModel returns a Model by id, optionally constrained by provider.
486// If provider is empty, the first matching id is returned. Looks up
487// against the merged active catalog.
488func FindModel(provider, id string) (Model, error) {
489 for _, m := range Active() {
490 if m.ID == id && (provider == "" || m.Provider == provider) {
491 return m, nil
492 }
493 }
494 return Model{}, fmt.Errorf("unknown model %q (provider=%q)", id, provider)
495}
496
497// ModelsForProvider returns all models for the given provider, from the
498// merged active catalog.
499func ModelsForProvider(provider string) []Model {
500 var out []Model
501 for _, m := range Active() {
502 if m.Provider == provider {
503 out = append(out, m)
504 }
505 }
506 return out
507}
508
509// ComputeCost returns the USD cost for the given usage on model m.
510func ComputeCost(m Model, u Usage) float64 {
511 inputPrice := m.PriceInput
512 outputPrice := m.PriceOutput
513 cacheReadPrice := m.PriceCacheRead
514 cacheWritePrice := m.PriceCacheWrite
515 promptTokens := u.InputTokens + u.CacheReadTokens + u.CacheWriteTokens
516 if m.PriceTierInputTokens > 0 && promptTokens > m.PriceTierInputTokens {
517 inputPrice = m.PriceInputAbove
518 outputPrice = m.PriceOutputAbove
519 cacheReadPrice = m.PriceCacheReadAbove
520 cacheWritePrice = m.PriceCacheWriteAbove
521 }
522
523 const per = 1_000_000.0
524 return float64(u.InputTokens)*inputPrice/per +
525 float64(u.OutputTokens)*outputPrice/per +
526 float64(u.CacheReadTokens)*cacheReadPrice/per +
527 float64(u.CacheWriteTokens)*cacheWritePrice/per
528}