@eidentic/model 0.1.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (7) hide show
  1. package/LICENSE +201 -0
  2. package/README.md +42 -0
  3. package/dist/index.cjs +1175 -0
  4. package/dist/index.d.cts +309 -0
  5. package/dist/index.d.ts +309 -0
  6. package/dist/index.js +1143 -0
  7. package/package.json +69 -0
package/dist/index.cjs ADDED
@@ -0,0 +1,1175 @@
1
+ "use strict";
2
+ var __defProp = Object.defineProperty;
3
+ var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
4
+ var __getOwnPropNames = Object.getOwnPropertyNames;
5
+ var __hasOwnProp = Object.prototype.hasOwnProperty;
6
+ var __export = (target, all) => {
7
+ for (var name in all)
8
+ __defProp(target, name, { get: all[name], enumerable: true });
9
+ };
10
+ var __copyProps = (to, from, except, desc) => {
11
+ if (from && typeof from === "object" || typeof from === "function") {
12
+ for (let key of __getOwnPropNames(from))
13
+ if (!__hasOwnProp.call(to, key) && key !== except)
14
+ __defProp(to, key, { get: () => from[key], enumerable: !(desc = __getOwnPropDesc(from, key)) || desc.enumerable });
15
+ }
16
+ return to;
17
+ };
18
+ var __toCommonJS = (mod) => __copyProps(__defProp({}, "__esModule", { value: true }), mod);
19
+
20
+ // src/index.ts
21
+ var index_exports = {};
22
+ __export(index_exports, {
23
+ AIEmbedder: () => AIEmbedder,
24
+ AIModel: () => AIModel,
25
+ LITELLM_URL: () => LITELLM_URL,
26
+ byTokenEstimate: () => byTokenEstimate,
27
+ cachedModel: () => cachedModel,
28
+ createOllamaModel: () => createOllamaModel,
29
+ defaultPrices: () => defaultPrices,
30
+ fetchLatestPrices: () => fetchLatestPrices,
31
+ mapLiteLLM: () => mapLiteLLM,
32
+ pricesUpdatedAt: () => pricesUpdatedAt,
33
+ routeModel: () => routeModel,
34
+ withFallback: () => withFallback
35
+ });
36
+ module.exports = __toCommonJS(index_exports);
37
+
38
+ // src/model.ts
39
+ var import_ai2 = require("ai");
40
+
41
+ // src/map.ts
42
+ var import_ai = require("ai");
43
+ var import_types = require("@eidentic/types");
44
+ var ANTHROPIC_CACHE_BREAKPOINT = {
45
+ anthropic: { cacheControl: { type: "ephemeral" } }
46
+ };
47
+ function mapMessages(msgs, opts) {
48
+ const systemParts = [];
49
+ const messages = [];
50
+ for (const m of msgs) {
51
+ if (m.role === "system") {
52
+ systemParts.push(typeof m.content === "string" ? m.content : textOf(m.content));
53
+ continue;
54
+ }
55
+ if (m.role === "user") {
56
+ const userBlocks = typeof m.content === "string" ? (0, import_types.decodeMultimodalInput)(m.content) : m.content;
57
+ if (userBlocks !== null) {
58
+ const hasImages = userBlocks.some((b) => b.type === "image");
59
+ if (hasImages) {
60
+ const parts = userBlocks.flatMap((b) => {
61
+ if (b.type === "text") return [{ type: "text", text: b.text }];
62
+ if (b.type === "image") {
63
+ const img = b.image;
64
+ const source = img.url ? new URL(img.url) : img.data;
65
+ if (source === void 0) return [];
66
+ const imagePart = {
67
+ type: "image",
68
+ image: source,
69
+ ...img.mediaType ? { mediaType: img.mediaType } : {}
70
+ };
71
+ return [imagePart];
72
+ }
73
+ return [];
74
+ });
75
+ if (parts.length > 0) messages.push({ role: "user", content: parts });
76
+ } else {
77
+ messages.push({ role: "user", content: textOf(userBlocks) });
78
+ }
79
+ } else {
80
+ messages.push({ role: "user", content: m.content });
81
+ }
82
+ continue;
83
+ }
84
+ if (m.role === "assistant") {
85
+ const parts = typeof m.content === "string" ? [{ type: "text", text: m.content }] : m.content.flatMap(
86
+ (b) => b.type === "text" ? [{ type: "text", text: b.text }] : b.type === "tool_use" ? [{ type: "tool-call", toolCallId: b.callId, toolName: b.name, input: b.input }] : []
87
+ );
88
+ if (parts.length > 0) messages.push({ role: "assistant", content: parts });
89
+ continue;
90
+ }
91
+ messages.push({
92
+ role: "tool",
93
+ content: [
94
+ {
95
+ type: "tool-result",
96
+ toolCallId: m.callId ?? "",
97
+ toolName: m.toolName ?? "",
98
+ output: { type: "text", value: typeof m.content === "string" ? m.content : textOf(m.content) }
99
+ }
100
+ ]
101
+ });
102
+ }
103
+ if (!systemParts.length) return { messages };
104
+ const systemText = systemParts.join("\n\n");
105
+ if (opts?.cacheControl) {
106
+ return {
107
+ system: { role: "system", content: systemText, providerOptions: ANTHROPIC_CACHE_BREAKPOINT },
108
+ messages
109
+ };
110
+ }
111
+ return { system: systemText, messages };
112
+ }
113
+ function textOf(content) {
114
+ return content.filter((b) => b.type === "text").map((b) => b.text).join("");
115
+ }
116
+ function buildTools(schemas) {
117
+ return Object.fromEntries(
118
+ schemas.map((s) => [
119
+ s.name,
120
+ (0, import_ai.tool)({ description: s.description, inputSchema: (0, import_ai.jsonSchema)(s.inputSchema) })
121
+ ])
122
+ );
123
+ }
124
+ function readStructuredOutput(result) {
125
+ try {
126
+ return result.experimental_output;
127
+ } catch {
128
+ return void 0;
129
+ }
130
+ }
131
+ function mapResult(result, opts) {
132
+ const content = [];
133
+ if (result.text) content.push({ type: "text", text: result.text });
134
+ for (const tc of result.toolCalls) {
135
+ content.push({ type: "tool_use", callId: tc.toolCallId, name: tc.toolName, input: tc.input });
136
+ }
137
+ const cached = result.usage.cachedInputTokens;
138
+ const object = opts?.structured && result.toolCalls.length === 0 ? readStructuredOutput(result) : void 0;
139
+ return {
140
+ content,
141
+ usage: {
142
+ inputTokens: result.usage.inputTokens ?? 0,
143
+ outputTokens: result.usage.outputTokens ?? 0,
144
+ ...cached !== void 0 && cached > 0 ? { cachedInputTokens: cached } : {}
145
+ },
146
+ ...object !== void 0 ? { object } : {}
147
+ };
148
+ }
149
+
150
+ // src/model.ts
151
+ var toolsArg = (t) => Object.keys(t).length > 0 ? t : void 0;
152
+ var outputArg = (schema) => schema !== void 0 ? import_ai2.Output.object({ schema: (0, import_ai2.jsonSchema)(schema) }) : void 0;
153
+ var AIModel = class {
154
+ resolve;
155
+ settings;
156
+ /** The model's own identifier, sourced from the AI SDK LanguageModel when a static model is passed. */
157
+ modelId;
158
+ constructor(model, options = {}) {
159
+ if (typeof model === "function") {
160
+ this.resolve = model;
161
+ this.modelId = void 0;
162
+ } else {
163
+ this.resolve = () => model;
164
+ this.modelId = model.modelId;
165
+ }
166
+ this.settings = options;
167
+ }
168
+ async complete(request) {
169
+ const model = await this.resolve(request.model);
170
+ const { system, messages } = mapMessages(request.messages, { cacheControl: request.cacheControl });
171
+ const tools = buildTools(request.tools);
172
+ const output = outputArg(request.outputSchema);
173
+ const result = await (0, import_ai2.generateText)({
174
+ model,
175
+ system,
176
+ messages,
177
+ tools: toolsArg(tools),
178
+ ...this.settings,
179
+ // no `stopWhen` and no tool `execute` → single round-trip; tool calls returned to our loop
180
+ ...output ? { experimental_output: output } : {},
181
+ ...request.signal ? { abortSignal: request.signal } : {}
182
+ });
183
+ return mapResult(result, { structured: output !== void 0 });
184
+ }
185
+ async *stream(request) {
186
+ const model = await this.resolve(request.model);
187
+ const { system, messages } = mapMessages(request.messages, { cacheControl: request.cacheControl });
188
+ const tools = buildTools(request.tools);
189
+ const output = outputArg(request.outputSchema);
190
+ const result = (0, import_ai2.streamText)({
191
+ model,
192
+ system,
193
+ messages,
194
+ tools: toolsArg(tools),
195
+ ...this.settings,
196
+ ...output ? { experimental_output: output } : {},
197
+ ...request.signal ? { abortSignal: request.signal } : {}
198
+ });
199
+ const toolUses = [];
200
+ for await (const part of result.fullStream) {
201
+ if (part.type === "text-delta") {
202
+ yield { type: "delta", delta: { text: part.text } };
203
+ } else if (part.type === "tool-call") {
204
+ toolUses.push({ type: "tool_use", callId: part.toolCallId, name: part.toolName, input: part.input });
205
+ } else if (part.type === "error") {
206
+ throw part.error instanceof Error ? part.error : new Error(`model stream error: ${String(part.error)}`);
207
+ }
208
+ }
209
+ if (request.signal?.aborted) {
210
+ throw new DOMException("stream aborted", "AbortError");
211
+ }
212
+ const text = await result.text;
213
+ const usage = await result.usage;
214
+ const content = [];
215
+ if (text) content.push({ type: "text", text });
216
+ content.push(...toolUses);
217
+ const cached = usage.cachedInputTokens;
218
+ let object;
219
+ if (output && toolUses.length === 0 && text) {
220
+ try {
221
+ object = JSON.parse(text);
222
+ } catch {
223
+ object = void 0;
224
+ }
225
+ }
226
+ yield {
227
+ type: "final",
228
+ response: {
229
+ content,
230
+ usage: {
231
+ inputTokens: usage.inputTokens ?? 0,
232
+ outputTokens: usage.outputTokens ?? 0,
233
+ ...cached !== void 0 && cached > 0 ? { cachedInputTokens: cached } : {}
234
+ },
235
+ ...object !== void 0 ? { object } : {}
236
+ }
237
+ };
238
+ }
239
+ };
240
+
241
+ // src/embedder.ts
242
+ var import_ai3 = require("ai");
243
+ var AIEmbedder = class _AIEmbedder {
244
+ constructor(model, dim) {
245
+ this.model = model;
246
+ this.dim = dim;
247
+ }
248
+ model;
249
+ dim;
250
+ /** Construct an embedder, probing the model once to discover its output dimension. */
251
+ static async create(model) {
252
+ const { embedding } = await (0, import_ai3.embed)({ model, value: "x" });
253
+ return new _AIEmbedder(model, embedding.length);
254
+ }
255
+ async embed(text) {
256
+ const { embedding } = await (0, import_ai3.embed)({ model: this.model, value: text });
257
+ return embedding;
258
+ }
259
+ /**
260
+ * Batch embedding via AI SDK v6 `embedMany({ model, values })` → `{ embeddings: number[][] }`.
261
+ * Embeds all texts in a single provider call (fewer round-trips on the ingest hot path).
262
+ * Each returned vector is validated to have length === `this.dim`.
263
+ */
264
+ async embedBatch(texts) {
265
+ if (texts.length === 0) return [];
266
+ const { embeddings } = await (0, import_ai3.embedMany)({ model: this.model, values: texts });
267
+ for (let i = 0; i < embeddings.length; i++) {
268
+ if (embeddings[i].length !== this.dim) {
269
+ throw new Error(`embedBatch: embedding[${i}] has length ${embeddings[i].length}, expected ${this.dim}`);
270
+ }
271
+ }
272
+ return embeddings;
273
+ }
274
+ };
275
+
276
+ // src/prices.ts
277
+ var pricesUpdatedAt = "2026-06-08T00:00:00.000Z";
278
+ var defaultPrices = {
279
+ "chatgpt-4o-latest": { inputPerMTok: 5, outputPerMTok: 15 },
280
+ "claude-3-5-haiku": { inputPerMTok: 1, outputPerMTok: 5 },
281
+ "claude-3-5-haiku@20241022": { inputPerMTok: 1, outputPerMTok: 5 },
282
+ "claude-3-5-sonnet": { inputPerMTok: 3, outputPerMTok: 15 },
283
+ "claude-3-5-sonnet@20240620": { inputPerMTok: 3, outputPerMTok: 15 },
284
+ "claude-3-7-sonnet-20250219": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
285
+ "claude-3-7-sonnet@20250219": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
286
+ "claude-3-haiku": { inputPerMTok: 0.25, outputPerMTok: 1.25 },
287
+ "claude-3-haiku-20240307": { inputPerMTok: 0.25, outputPerMTok: 1.25, cachedInputPerMTok: 0.03 },
288
+ "claude-3-haiku@20240307": { inputPerMTok: 0.25, outputPerMTok: 1.25 },
289
+ "claude-3-opus": { inputPerMTok: 15, outputPerMTok: 75 },
290
+ "claude-3-opus-20240229": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
291
+ "claude-3-opus@20240229": { inputPerMTok: 15, outputPerMTok: 75 },
292
+ "claude-3-sonnet": { inputPerMTok: 3, outputPerMTok: 15 },
293
+ "claude-3-sonnet@20240229": { inputPerMTok: 3, outputPerMTok: 15 },
294
+ "claude-4-opus-20250514": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
295
+ "claude-4-sonnet-20250514": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
296
+ "claude-haiku-4-5": { inputPerMTok: 1, outputPerMTok: 5, cachedInputPerMTok: 0.1 },
297
+ "claude-haiku-4-5-20251001": { inputPerMTok: 1, outputPerMTok: 5, cachedInputPerMTok: 0.1 },
298
+ "claude-haiku-4-5@20251001": { inputPerMTok: 1, outputPerMTok: 5, cachedInputPerMTok: 0.1 },
299
+ "claude-opus-4": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
300
+ "claude-opus-4-1": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
301
+ "claude-opus-4-1-20250805": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
302
+ "claude-opus-4-1@20250805": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
303
+ "claude-opus-4-20250514": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
304
+ "claude-opus-4-5": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
305
+ "claude-opus-4-5-20251101": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
306
+ "claude-opus-4-5@20251101": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
307
+ "claude-opus-4-6": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
308
+ "claude-opus-4-6-20260205": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
309
+ "claude-opus-4-6@default": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
310
+ "claude-opus-4-7": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
311
+ "claude-opus-4-7-20260416": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
312
+ "claude-opus-4-7@default": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
313
+ "claude-opus-4-8": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
314
+ "claude-opus-4-8@default": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
315
+ "claude-opus-4@20250514": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
316
+ "claude-sonnet-4": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
317
+ "claude-sonnet-4-20250514": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
318
+ "claude-sonnet-4-5": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
319
+ "claude-sonnet-4-5-20250929": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
320
+ "claude-sonnet-4-5@20250929": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
321
+ "claude-sonnet-4-6": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
322
+ "claude-sonnet-4-6@default": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
323
+ "claude-sonnet-4@20250514": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
324
+ "codestral-2": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
325
+ "codestral-2405": { inputPerMTok: 1, outputPerMTok: 3 },
326
+ "codestral-2501": { inputPerMTok: 0.2, outputPerMTok: 0.6 },
327
+ "codestral-2508": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
328
+ "codestral-2@001": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
329
+ "codestral-latest": { inputPerMTok: 1, outputPerMTok: 3 },
330
+ "codestral-mamba-latest": { inputPerMTok: 0.25, outputPerMTok: 0.25 },
331
+ "codestral@2405": { inputPerMTok: 0.2, outputPerMTok: 0.6 },
332
+ "codestral@latest": { inputPerMTok: 0.2, outputPerMTok: 0.6 },
333
+ "codex-mini-latest": { inputPerMTok: 1.5, outputPerMTok: 6, cachedInputPerMTok: 0.375 },
334
+ "cohere/embed-v4.0": { inputPerMTok: 0.12, outputPerMTok: 0 },
335
+ "command": { inputPerMTok: 1, outputPerMTok: 2 },
336
+ "command-nightly": { inputPerMTok: 1, outputPerMTok: 2 },
337
+ "deep-research-pro-preview-12-2025": { inputPerMTok: 2, outputPerMTok: 12 },
338
+ "deepseek-chat": { inputPerMTok: 0.28, outputPerMTok: 0.42, cachedInputPerMTok: 0.028 },
339
+ "deepseek-coder": { inputPerMTok: 0.14, outputPerMTok: 0.28 },
340
+ "deepseek-r1": { inputPerMTok: 0.55, outputPerMTok: 2.19 },
341
+ "deepseek-reasoner": { inputPerMTok: 0.28, outputPerMTok: 0.42, cachedInputPerMTok: 0.028 },
342
+ "deepseek-v3": { inputPerMTok: 0.27, outputPerMTok: 1.1, cachedInputPerMTok: 0.07 },
343
+ "deepseek-v3.2": { inputPerMTok: 0.28, outputPerMTok: 0.4 },
344
+ "deepseek/deepseek-chat": { inputPerMTok: 0.28, outputPerMTok: 0.42, cachedInputPerMTok: 0.028 },
345
+ "deepseek/deepseek-coder": { inputPerMTok: 0.14, outputPerMTok: 0.28 },
346
+ "deepseek/deepseek-r1": { inputPerMTok: 0.55, outputPerMTok: 2.19 },
347
+ "deepseek/deepseek-reasoner": { inputPerMTok: 0.28, outputPerMTok: 0.42, cachedInputPerMTok: 0.028 },
348
+ "deepseek/deepseek-v3": { inputPerMTok: 0.27, outputPerMTok: 1.1, cachedInputPerMTok: 0.07 },
349
+ "deepseek/deepseek-v3.2": { inputPerMTok: 0.28, outputPerMTok: 0.4 },
350
+ "devstral-2512": { inputPerMTok: 0.4, outputPerMTok: 2 },
351
+ "devstral-latest": { inputPerMTok: 0.4, outputPerMTok: 2 },
352
+ "devstral-medium-2507": { inputPerMTok: 0.4, outputPerMTok: 2 },
353
+ "devstral-medium-latest": { inputPerMTok: 0.4, outputPerMTok: 2 },
354
+ "devstral-small-2505": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
355
+ "devstral-small-2507": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
356
+ "devstral-small-latest": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
357
+ "embed-english-light-v2.0": { inputPerMTok: 0.1, outputPerMTok: 0 },
358
+ "embed-english-light-v3.0": { inputPerMTok: 0.1, outputPerMTok: 0 },
359
+ "embed-english-v2.0": { inputPerMTok: 0.1, outputPerMTok: 0 },
360
+ "embed-english-v3.0": { inputPerMTok: 0.1, outputPerMTok: 0 },
361
+ "embed-multilingual-light-v3.0": { inputPerMTok: 100, outputPerMTok: 0 },
362
+ "embed-multilingual-v2.0": { inputPerMTok: 0.1, outputPerMTok: 0 },
363
+ "embed-multilingual-v3.0": { inputPerMTok: 0.1, outputPerMTok: 0 },
364
+ "embed-v4.0": { inputPerMTok: 0.12, outputPerMTok: 0 },
365
+ "ft:gpt-3.5-turbo": { inputPerMTok: 3, outputPerMTok: 6 },
366
+ "ft:gpt-3.5-turbo-0125": { inputPerMTok: 3, outputPerMTok: 6 },
367
+ "ft:gpt-3.5-turbo-0613": { inputPerMTok: 3, outputPerMTok: 6 },
368
+ "ft:gpt-3.5-turbo-1106": { inputPerMTok: 3, outputPerMTok: 6 },
369
+ "ft:gpt-4-0613": { inputPerMTok: 30, outputPerMTok: 60 },
370
+ "ft:gpt-4.1-2025-04-14": { inputPerMTok: 3, outputPerMTok: 12, cachedInputPerMTok: 0.75 },
371
+ "ft:gpt-4.1-mini-2025-04-14": { inputPerMTok: 0.8, outputPerMTok: 3.2, cachedInputPerMTok: 0.2 },
372
+ "ft:gpt-4.1-nano-2025-04-14": { inputPerMTok: 0.2, outputPerMTok: 0.8, cachedInputPerMTok: 0.05 },
373
+ "ft:gpt-4o-2024-08-06": { inputPerMTok: 3.75, outputPerMTok: 15, cachedInputPerMTok: 1.875 },
374
+ "ft:gpt-4o-2024-11-20": { inputPerMTok: 3.75, outputPerMTok: 15 },
375
+ "ft:gpt-4o-mini-2024-07-18": { inputPerMTok: 0.3, outputPerMTok: 1.2, cachedInputPerMTok: 0.15 },
376
+ "ft:o4-mini-2025-04-16": { inputPerMTok: 4, outputPerMTok: 16, cachedInputPerMTok: 1 },
377
+ "gemini-1.5-flash": { inputPerMTok: 0.075, outputPerMTok: 0 },
378
+ "gemini-2.0-flash": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.025 },
379
+ "gemini-2.0-flash-001": { inputPerMTok: 0.15, outputPerMTok: 0.6, cachedInputPerMTok: 0.0375 },
380
+ "gemini-2.0-flash-exp-image-generation": { inputPerMTok: 0, outputPerMTok: 0 },
381
+ "gemini-2.0-flash-lite": { inputPerMTok: 0.075, outputPerMTok: 0.3, cachedInputPerMTok: 0.01875 },
382
+ "gemini-2.0-flash-lite-001": { inputPerMTok: 0.075, outputPerMTok: 0.3, cachedInputPerMTok: 0.01875 },
383
+ "gemini-2.5-computer-use-preview-10-2025": { inputPerMTok: 1.25, outputPerMTok: 10 },
384
+ "gemini-2.5-flash": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.03 },
385
+ "gemini-2.5-flash-image": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.03 },
386
+ "gemini-2.5-flash-lite": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.01 },
387
+ "gemini-2.5-flash-lite-preview-06-17": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.025 },
388
+ "gemini-2.5-flash-lite-preview-09-2025": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.01 },
389
+ "gemini-2.5-flash-native-audio-latest": { inputPerMTok: 0.3, outputPerMTok: 2.5 },
390
+ "gemini-2.5-flash-native-audio-preview-09-2025": { inputPerMTok: 0.3, outputPerMTok: 2.5 },
391
+ "gemini-2.5-flash-native-audio-preview-12-2025": { inputPerMTok: 0.3, outputPerMTok: 2.5 },
392
+ "gemini-2.5-flash-preview-09-2025": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.075 },
393
+ "gemini-2.5-flash-preview-tts": { inputPerMTok: 0.3, outputPerMTok: 2.5 },
394
+ "gemini-2.5-pro": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
395
+ "gemini-2.5-pro-preview-tts": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
396
+ "gemini-3-flash-preview": { inputPerMTok: 0.5, outputPerMTok: 3, cachedInputPerMTok: 0.05 },
397
+ "gemini-3-pro-image-preview": { inputPerMTok: 2, outputPerMTok: 12 },
398
+ "gemini-3-pro-preview": { inputPerMTok: 2, outputPerMTok: 12, cachedInputPerMTok: 0.2 },
399
+ "gemini-3.1-flash-image-preview": { inputPerMTok: 0.5, outputPerMTok: 3 },
400
+ "gemini-3.1-flash-lite": { inputPerMTok: 0.25, outputPerMTok: 1.5, cachedInputPerMTok: 0.025 },
401
+ "gemini-3.1-flash-lite-preview": { inputPerMTok: 0.25, outputPerMTok: 1.5, cachedInputPerMTok: 0.025 },
402
+ "gemini-3.1-flash-live-preview": { inputPerMTok: 0.75, outputPerMTok: 4.5 },
403
+ "gemini-3.1-pro-preview": { inputPerMTok: 2, outputPerMTok: 12, cachedInputPerMTok: 0.2 },
404
+ "gemini-3.1-pro-preview-customtools": { inputPerMTok: 2, outputPerMTok: 12, cachedInputPerMTok: 0.2 },
405
+ "gemini-3.5-flash": { inputPerMTok: 1.5, outputPerMTok: 9, cachedInputPerMTok: 0.15 },
406
+ "gemini-embedding-001": { inputPerMTok: 0.15, outputPerMTok: 0 },
407
+ "gemini-embedding-2": { inputPerMTok: 0.2, outputPerMTok: 0 },
408
+ "gemini-embedding-2-preview": { inputPerMTok: 0.2, outputPerMTok: 0 },
409
+ "gemini-exp-1114": { inputPerMTok: 0, outputPerMTok: 0 },
410
+ "gemini-exp-1206": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.03 },
411
+ "gemini-flash-experimental": { inputPerMTok: 0, outputPerMTok: 0 },
412
+ "gemini-flash-latest": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.03 },
413
+ "gemini-flash-lite-latest": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.01 },
414
+ "gemini-gemma-2-27b-it": { inputPerMTok: 0.35, outputPerMTok: 1.05 },
415
+ "gemini-gemma-2-9b-it": { inputPerMTok: 0.35, outputPerMTok: 1.05 },
416
+ "gemini-live-2.5-flash-preview-native-audio-09-2025": { inputPerMTok: 0.3, outputPerMTok: 2, cachedInputPerMTok: 0.075 },
417
+ "gemini-pro-latest": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
418
+ "gemini-robotics-er-1.5-preview": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0 },
419
+ "gemini/deep-research-pro-preview-12-2025": { inputPerMTok: 2, outputPerMTok: 12 },
420
+ "gemini/gemini-1.5-flash": { inputPerMTok: 0.075, outputPerMTok: 0 },
421
+ "gemini/gemini-2.0-flash": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.025 },
422
+ "gemini/gemini-2.0-flash-001": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.025 },
423
+ "gemini/gemini-2.0-flash-exp-image-generation": { inputPerMTok: 0, outputPerMTok: 0 },
424
+ "gemini/gemini-2.0-flash-lite": { inputPerMTok: 0.075, outputPerMTok: 0.3, cachedInputPerMTok: 0.01875 },
425
+ "gemini/gemini-2.0-flash-lite-001": { inputPerMTok: 0.075, outputPerMTok: 0.3, cachedInputPerMTok: 0.01875 },
426
+ "gemini/gemini-2.5-computer-use-preview-10-2025": { inputPerMTok: 1.25, outputPerMTok: 10 },
427
+ "gemini/gemini-2.5-flash": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.03 },
428
+ "gemini/gemini-2.5-flash-image": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.03 },
429
+ "gemini/gemini-2.5-flash-lite": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.01 },
430
+ "gemini/gemini-2.5-flash-lite-preview-06-17": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.025 },
431
+ "gemini/gemini-2.5-flash-lite-preview-09-2025": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.01 },
432
+ "gemini/gemini-2.5-flash-native-audio-latest": { inputPerMTok: 0.3, outputPerMTok: 2.5 },
433
+ "gemini/gemini-2.5-flash-native-audio-preview-09-2025": { inputPerMTok: 0.3, outputPerMTok: 2.5 },
434
+ "gemini/gemini-2.5-flash-native-audio-preview-12-2025": { inputPerMTok: 0.3, outputPerMTok: 2.5 },
435
+ "gemini/gemini-2.5-flash-preview-09-2025": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.075 },
436
+ "gemini/gemini-2.5-flash-preview-tts": { inputPerMTok: 0.3, outputPerMTok: 2.5 },
437
+ "gemini/gemini-2.5-pro": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
438
+ "gemini/gemini-2.5-pro-preview-tts": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
439
+ "gemini/gemini-3-flash-preview": { inputPerMTok: 0.5, outputPerMTok: 3, cachedInputPerMTok: 0.05 },
440
+ "gemini/gemini-3-pro-image-preview": { inputPerMTok: 2, outputPerMTok: 12 },
441
+ "gemini/gemini-3-pro-preview": { inputPerMTok: 2, outputPerMTok: 12, cachedInputPerMTok: 0.2 },
442
+ "gemini/gemini-3.1-flash-image-preview": { inputPerMTok: 0.25, outputPerMTok: 1.5 },
443
+ "gemini/gemini-3.1-flash-lite": { inputPerMTok: 0.25, outputPerMTok: 1.5, cachedInputPerMTok: 0.025 },
444
+ "gemini/gemini-3.1-flash-lite-preview": { inputPerMTok: 0.25, outputPerMTok: 1.5, cachedInputPerMTok: 0.025 },
445
+ "gemini/gemini-3.1-flash-live-preview": { inputPerMTok: 0.75, outputPerMTok: 4.5 },
446
+ "gemini/gemini-3.1-pro-preview": { inputPerMTok: 2, outputPerMTok: 12, cachedInputPerMTok: 0.2 },
447
+ "gemini/gemini-3.1-pro-preview-customtools": { inputPerMTok: 2, outputPerMTok: 12, cachedInputPerMTok: 0.2 },
448
+ "gemini/gemini-3.5-flash": { inputPerMTok: 1.5, outputPerMTok: 9, cachedInputPerMTok: 0.15 },
449
+ "gemini/gemini-embedding-001": { inputPerMTok: 0.15, outputPerMTok: 0 },
450
+ "gemini/gemini-embedding-2": { inputPerMTok: 0.2, outputPerMTok: 0 },
451
+ "gemini/gemini-embedding-2-preview": { inputPerMTok: 0.2, outputPerMTok: 0 },
452
+ "gemini/gemini-exp-1114": { inputPerMTok: 0, outputPerMTok: 0 },
453
+ "gemini/gemini-exp-1206": { inputPerMTok: 0, outputPerMTok: 0 },
454
+ "gemini/gemini-flash-latest": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.075 },
455
+ "gemini/gemini-flash-lite-latest": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.025 },
456
+ "gemini/gemini-gemma-2-27b-it": { inputPerMTok: 0.35, outputPerMTok: 1.05 },
457
+ "gemini/gemini-gemma-2-9b-it": { inputPerMTok: 0.35, outputPerMTok: 1.05 },
458
+ "gemini/gemini-live-2.5-flash-preview-native-audio-09-2025": { inputPerMTok: 0.3, outputPerMTok: 2, cachedInputPerMTok: 0.075 },
459
+ "gemini/gemini-pro-latest": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
460
+ "gemini/gemini-robotics-er-1.5-preview": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0 },
461
+ "gemini/gemma-3-27b-it": { inputPerMTok: 0, outputPerMTok: 0 },
462
+ "gemini/learnlm-1.5-pro-experimental": { inputPerMTok: 0, outputPerMTok: 0 },
463
+ "gemini/lyria-3-clip-preview": { inputPerMTok: 0, outputPerMTok: 0 },
464
+ "gemini/lyria-3-pro-preview": { inputPerMTok: 0, outputPerMTok: 0 },
465
+ "gemma-3-27b-it": { inputPerMTok: 0, outputPerMTok: 0 },
466
+ "gpt-3.5-turbo": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
467
+ "gpt-3.5-turbo-0125": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
468
+ "gpt-3.5-turbo-1106": { inputPerMTok: 1, outputPerMTok: 2 },
469
+ "gpt-3.5-turbo-16k": { inputPerMTok: 3, outputPerMTok: 4 },
470
+ "gpt-4": { inputPerMTok: 30, outputPerMTok: 60 },
471
+ "gpt-4-0125-preview": { inputPerMTok: 10, outputPerMTok: 30 },
472
+ "gpt-4-0314": { inputPerMTok: 30, outputPerMTok: 60 },
473
+ "gpt-4-0613": { inputPerMTok: 30, outputPerMTok: 60 },
474
+ "gpt-4-1106-preview": { inputPerMTok: 10, outputPerMTok: 30 },
475
+ "gpt-4-turbo": { inputPerMTok: 10, outputPerMTok: 30 },
476
+ "gpt-4-turbo-2024-04-09": { inputPerMTok: 10, outputPerMTok: 30 },
477
+ "gpt-4-turbo-preview": { inputPerMTok: 10, outputPerMTok: 30 },
478
+ "gpt-4.1": { inputPerMTok: 2, outputPerMTok: 8, cachedInputPerMTok: 0.5 },
479
+ "gpt-4.1-2025-04-14": { inputPerMTok: 2, outputPerMTok: 8, cachedInputPerMTok: 0.5 },
480
+ "gpt-4.1-mini": { inputPerMTok: 0.4, outputPerMTok: 1.6, cachedInputPerMTok: 0.1 },
481
+ "gpt-4.1-mini-2025-04-14": { inputPerMTok: 0.4, outputPerMTok: 1.6, cachedInputPerMTok: 0.1 },
482
+ "gpt-4.1-nano": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.025 },
483
+ "gpt-4.1-nano-2025-04-14": { inputPerMTok: 0.1, outputPerMTok: 0.4, cachedInputPerMTok: 0.025 },
484
+ "gpt-4o": { inputPerMTok: 2.5, outputPerMTok: 10, cachedInputPerMTok: 1.25 },
485
+ "gpt-4o-2024-05-13": { inputPerMTok: 5, outputPerMTok: 15 },
486
+ "gpt-4o-2024-08-06": { inputPerMTok: 2.5, outputPerMTok: 10, cachedInputPerMTok: 1.25 },
487
+ "gpt-4o-2024-11-20": { inputPerMTok: 2.5, outputPerMTok: 10, cachedInputPerMTok: 1.25 },
488
+ "gpt-4o-audio-preview": { inputPerMTok: 2.5, outputPerMTok: 10 },
489
+ "gpt-4o-audio-preview-2024-12-17": { inputPerMTok: 2.5, outputPerMTok: 10 },
490
+ "gpt-4o-audio-preview-2025-06-03": { inputPerMTok: 2.5, outputPerMTok: 10 },
491
+ "gpt-4o-mini": { inputPerMTok: 0.15, outputPerMTok: 0.6, cachedInputPerMTok: 0.075 },
492
+ "gpt-4o-mini-2024-07-18": { inputPerMTok: 0.15, outputPerMTok: 0.6, cachedInputPerMTok: 0.075 },
493
+ "gpt-4o-mini-audio-preview": { inputPerMTok: 0.15, outputPerMTok: 0.6 },
494
+ "gpt-4o-mini-audio-preview-2024-12-17": { inputPerMTok: 0.15, outputPerMTok: 0.6 },
495
+ "gpt-4o-mini-realtime-preview": { inputPerMTok: 0.6, outputPerMTok: 2.4, cachedInputPerMTok: 0.3 },
496
+ "gpt-4o-mini-realtime-preview-2024-12-17": { inputPerMTok: 0.6, outputPerMTok: 2.4, cachedInputPerMTok: 0.3 },
497
+ "gpt-4o-mini-search-preview": { inputPerMTok: 0.15, outputPerMTok: 0.6, cachedInputPerMTok: 0.075 },
498
+ "gpt-4o-mini-search-preview-2025-03-11": { inputPerMTok: 0.15, outputPerMTok: 0.6, cachedInputPerMTok: 0.075 },
499
+ "gpt-4o-mini-transcribe": { inputPerMTok: 1.25, outputPerMTok: 5 },
500
+ "gpt-4o-mini-transcribe-2025-03-20": { inputPerMTok: 1.25, outputPerMTok: 5 },
501
+ "gpt-4o-mini-transcribe-2025-12-15": { inputPerMTok: 1.25, outputPerMTok: 5 },
502
+ "gpt-4o-mini-tts": { inputPerMTok: 2.5, outputPerMTok: 10 },
503
+ "gpt-4o-mini-tts-2025-03-20": { inputPerMTok: 2.5, outputPerMTok: 10 },
504
+ "gpt-4o-mini-tts-2025-12-15": { inputPerMTok: 2.5, outputPerMTok: 10 },
505
+ "gpt-4o-realtime-preview": { inputPerMTok: 5, outputPerMTok: 20, cachedInputPerMTok: 2.5 },
506
+ "gpt-4o-realtime-preview-2024-12-17": { inputPerMTok: 5, outputPerMTok: 20, cachedInputPerMTok: 2.5 },
507
+ "gpt-4o-realtime-preview-2025-06-03": { inputPerMTok: 5, outputPerMTok: 20, cachedInputPerMTok: 2.5 },
508
+ "gpt-4o-search-preview": { inputPerMTok: 2.5, outputPerMTok: 10, cachedInputPerMTok: 1.25 },
509
+ "gpt-4o-search-preview-2025-03-11": { inputPerMTok: 2.5, outputPerMTok: 10, cachedInputPerMTok: 1.25 },
510
+ "gpt-4o-transcribe": { inputPerMTok: 2.5, outputPerMTok: 10 },
511
+ "gpt-4o-transcribe-diarize": { inputPerMTok: 2.5, outputPerMTok: 10 },
512
+ "gpt-5": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
513
+ "gpt-5-2025-08-07": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
514
+ "gpt-5-chat": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
515
+ "gpt-5-chat-latest": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
516
+ "gpt-5-codex": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
517
+ "gpt-5-mini": { inputPerMTok: 0.25, outputPerMTok: 2, cachedInputPerMTok: 0.025 },
518
+ "gpt-5-mini-2025-08-07": { inputPerMTok: 0.25, outputPerMTok: 2, cachedInputPerMTok: 0.025 },
519
+ "gpt-5-nano": { inputPerMTok: 0.05, outputPerMTok: 0.4, cachedInputPerMTok: 5e-3 },
520
+ "gpt-5-nano-2025-08-07": { inputPerMTok: 0.05, outputPerMTok: 0.4, cachedInputPerMTok: 5e-3 },
521
+ "gpt-5-pro": { inputPerMTok: 15, outputPerMTok: 120 },
522
+ "gpt-5-pro-2025-10-06": { inputPerMTok: 15, outputPerMTok: 120 },
523
+ "gpt-5-search-api": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
524
+ "gpt-5-search-api-2025-10-14": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
525
+ "gpt-5.1": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
526
+ "gpt-5.1-2025-11-13": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
527
+ "gpt-5.1-chat-latest": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
528
+ "gpt-5.1-codex": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
529
+ "gpt-5.1-codex-max": { inputPerMTok: 1.25, outputPerMTok: 10, cachedInputPerMTok: 0.125 },
530
+ "gpt-5.1-codex-mini": { inputPerMTok: 0.25, outputPerMTok: 2, cachedInputPerMTok: 0.025 },
531
+ "gpt-5.2": { inputPerMTok: 1.75, outputPerMTok: 14, cachedInputPerMTok: 0.175 },
532
+ "gpt-5.2-2025-12-11": { inputPerMTok: 1.75, outputPerMTok: 14, cachedInputPerMTok: 0.175 },
533
+ "gpt-5.2-chat-latest": { inputPerMTok: 1.75, outputPerMTok: 14, cachedInputPerMTok: 0.175 },
534
+ "gpt-5.2-codex": { inputPerMTok: 1.75, outputPerMTok: 14, cachedInputPerMTok: 0.175 },
535
+ "gpt-5.2-pro": { inputPerMTok: 21, outputPerMTok: 168 },
536
+ "gpt-5.2-pro-2025-12-11": { inputPerMTok: 21, outputPerMTok: 168 },
537
+ "gpt-5.3-chat-latest": { inputPerMTok: 1.75, outputPerMTok: 14, cachedInputPerMTok: 0.175 },
538
+ "gpt-5.3-codex": { inputPerMTok: 1.75, outputPerMTok: 14, cachedInputPerMTok: 0.175 },
539
+ "gpt-5.4": { inputPerMTok: 2.5, outputPerMTok: 15, cachedInputPerMTok: 0.25 },
540
+ "gpt-5.4-2026-03-05": { inputPerMTok: 2.5, outputPerMTok: 15, cachedInputPerMTok: 0.25 },
541
+ "gpt-5.4-mini": { inputPerMTok: 0.75, outputPerMTok: 4.5, cachedInputPerMTok: 0.075 },
542
+ "gpt-5.4-mini-2026-03-17": { inputPerMTok: 0.75, outputPerMTok: 4.5, cachedInputPerMTok: 0.075 },
543
+ "gpt-5.4-nano": { inputPerMTok: 0.2, outputPerMTok: 1.25, cachedInputPerMTok: 0.02 },
544
+ "gpt-5.4-nano-2026-03-17": { inputPerMTok: 0.2, outputPerMTok: 1.25, cachedInputPerMTok: 0.02 },
545
+ "gpt-5.4-pro": { inputPerMTok: 30, outputPerMTok: 180, cachedInputPerMTok: 3 },
546
+ "gpt-5.4-pro-2026-03-05": { inputPerMTok: 30, outputPerMTok: 180, cachedInputPerMTok: 3 },
547
+ "gpt-5.5": { inputPerMTok: 5, outputPerMTok: 30, cachedInputPerMTok: 0.5 },
548
+ "gpt-5.5-2026-04-23": { inputPerMTok: 5, outputPerMTok: 30, cachedInputPerMTok: 0.5 },
549
+ "gpt-5.5-pro": { inputPerMTok: 30, outputPerMTok: 180, cachedInputPerMTok: 3 },
550
+ "gpt-5.5-pro-2026-04-23": { inputPerMTok: 30, outputPerMTok: 180, cachedInputPerMTok: 3 },
551
+ "gpt-audio": { inputPerMTok: 2.5, outputPerMTok: 10 },
552
+ "gpt-audio-1.5": { inputPerMTok: 2.5, outputPerMTok: 10 },
553
+ "gpt-audio-2025-08-28": { inputPerMTok: 2.5, outputPerMTok: 10 },
554
+ "gpt-audio-mini": { inputPerMTok: 0.6, outputPerMTok: 2.4 },
555
+ "gpt-audio-mini-2025-10-06": { inputPerMTok: 0.6, outputPerMTok: 2.4 },
556
+ "gpt-audio-mini-2025-12-15": { inputPerMTok: 0.6, outputPerMTok: 2.4 },
557
+ "gpt-image-1.5": { inputPerMTok: 5, outputPerMTok: 10, cachedInputPerMTok: 1.25 },
558
+ "gpt-image-1.5-2025-12-16": { inputPerMTok: 5, outputPerMTok: 10, cachedInputPerMTok: 1.25 },
559
+ "gpt-image-2": { inputPerMTok: 5, outputPerMTok: 10, cachedInputPerMTok: 1.25 },
560
+ "gpt-image-2-2026-04-21": { inputPerMTok: 5, outputPerMTok: 10, cachedInputPerMTok: 1.25 },
561
+ "gpt-realtime": { inputPerMTok: 4, outputPerMTok: 16, cachedInputPerMTok: 0.4 },
562
+ "gpt-realtime-1.5": { inputPerMTok: 4, outputPerMTok: 16, cachedInputPerMTok: 0.4 },
563
+ "gpt-realtime-2": { inputPerMTok: 4, outputPerMTok: 16, cachedInputPerMTok: 0.4 },
564
+ "gpt-realtime-2025-08-28": { inputPerMTok: 4, outputPerMTok: 16, cachedInputPerMTok: 0.4 },
565
+ "gpt-realtime-mini": { inputPerMTok: 0.6, outputPerMTok: 2.4 },
566
+ "gpt-realtime-mini-2025-10-06": { inputPerMTok: 0.6, outputPerMTok: 2.4, cachedInputPerMTok: 0.06 },
567
+ "gpt-realtime-mini-2025-12-15": { inputPerMTok: 0.6, outputPerMTok: 2.4, cachedInputPerMTok: 0.06 },
568
+ "grok-2": { inputPerMTok: 2, outputPerMTok: 10 },
569
+ "grok-2-1212": { inputPerMTok: 2, outputPerMTok: 10 },
570
+ "grok-2-latest": { inputPerMTok: 2, outputPerMTok: 10 },
571
+ "grok-2-vision": { inputPerMTok: 2, outputPerMTok: 10 },
572
+ "grok-2-vision-1212": { inputPerMTok: 2, outputPerMTok: 10 },
573
+ "grok-2-vision-latest": { inputPerMTok: 2, outputPerMTok: 10 },
574
+ "grok-3": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.75 },
575
+ "grok-3-beta": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.75 },
576
+ "grok-3-fast-beta": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 1.25 },
577
+ "grok-3-fast-latest": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 1.25 },
578
+ "grok-3-latest": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.75 },
579
+ "grok-3-mini": { inputPerMTok: 0.3, outputPerMTok: 0.5, cachedInputPerMTok: 0.075 },
580
+ "grok-3-mini-beta": { inputPerMTok: 0.3, outputPerMTok: 0.5, cachedInputPerMTok: 0.075 },
581
+ "grok-3-mini-fast": { inputPerMTok: 0.6, outputPerMTok: 4, cachedInputPerMTok: 0.15 },
582
+ "grok-3-mini-fast-beta": { inputPerMTok: 0.6, outputPerMTok: 4, cachedInputPerMTok: 0.15 },
583
+ "grok-3-mini-fast-latest": { inputPerMTok: 0.6, outputPerMTok: 4, cachedInputPerMTok: 0.15 },
584
+ "grok-3-mini-latest": { inputPerMTok: 0.3, outputPerMTok: 0.5, cachedInputPerMTok: 0.075 },
585
+ "grok-4": { inputPerMTok: 3, outputPerMTok: 15 },
586
+ "grok-4-0709": { inputPerMTok: 3, outputPerMTok: 15 },
587
+ "grok-4-1-fast": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
588
+ "grok-4-1-fast-non-reasoning": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
589
+ "grok-4-1-fast-non-reasoning-latest": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
590
+ "grok-4-1-fast-reasoning": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
591
+ "grok-4-1-fast-reasoning-latest": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
592
+ "grok-4-fast-non-reasoning": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
593
+ "grok-4-fast-reasoning": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
594
+ "grok-4-latest": { inputPerMTok: 3, outputPerMTok: 15 },
595
+ "grok-4.20-0309-reasoning": { inputPerMTok: 2, outputPerMTok: 6, cachedInputPerMTok: 0.2 },
596
+ "grok-4.20-beta-0309-non-reasoning": { inputPerMTok: 2, outputPerMTok: 6, cachedInputPerMTok: 0.2 },
597
+ "grok-4.20-beta-0309-reasoning": { inputPerMTok: 2, outputPerMTok: 6, cachedInputPerMTok: 0.2 },
598
+ "grok-4.20-multi-agent-beta-0309": { inputPerMTok: 2, outputPerMTok: 6, cachedInputPerMTok: 0.2 },
599
+ "grok-4.3": { inputPerMTok: 1.25, outputPerMTok: 2.5, cachedInputPerMTok: 0.2 },
600
+ "grok-4.3-latest": { inputPerMTok: 1.25, outputPerMTok: 2.5, cachedInputPerMTok: 0.2 },
601
+ "grok-beta": { inputPerMTok: 5, outputPerMTok: 15 },
602
+ "grok-code-fast": { inputPerMTok: 0.2, outputPerMTok: 1.5, cachedInputPerMTok: 0.02 },
603
+ "grok-code-fast-1": { inputPerMTok: 0.2, outputPerMTok: 1.5, cachedInputPerMTok: 0.02 },
604
+ "grok-code-fast-1-0825": { inputPerMTok: 0.2, outputPerMTok: 1.5, cachedInputPerMTok: 0.02 },
605
+ "grok-vision-beta": { inputPerMTok: 5, outputPerMTok: 15 },
606
+ "labs-devstral-small-2512": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
607
+ "learnlm-1.5-pro-experimental": { inputPerMTok: 0, outputPerMTok: 0 },
608
+ "lyria-3-clip-preview": { inputPerMTok: 0, outputPerMTok: 0 },
609
+ "lyria-3-pro-preview": { inputPerMTok: 0, outputPerMTok: 0 },
610
+ "magistral-medium-1-2-2509": { inputPerMTok: 2, outputPerMTok: 5 },
611
+ "magistral-medium-2506": { inputPerMTok: 2, outputPerMTok: 5 },
612
+ "magistral-medium-2509": { inputPerMTok: 2, outputPerMTok: 5 },
613
+ "magistral-medium-latest": { inputPerMTok: 2, outputPerMTok: 5 },
614
+ "magistral-small-1-2-2509": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
615
+ "magistral-small-2506": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
616
+ "magistral-small-latest": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
617
+ "ministral-3-14b-2512": { inputPerMTok: 0.2, outputPerMTok: 0.2 },
618
+ "ministral-3-3b-2512": { inputPerMTok: 0.1, outputPerMTok: 0.1 },
619
+ "ministral-3-8b-2512": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
620
+ "ministral-8b-2512": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
621
+ "ministral-8b-latest": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
622
+ "mistral-large-2402": { inputPerMTok: 4, outputPerMTok: 12 },
623
+ "mistral-large-2407": { inputPerMTok: 3, outputPerMTok: 9 },
624
+ "mistral-large-2411": { inputPerMTok: 2, outputPerMTok: 6 },
625
+ "mistral-large-2512": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
626
+ "mistral-large-3": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
627
+ "mistral-large-latest": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
628
+ "mistral-large@2407": { inputPerMTok: 2, outputPerMTok: 6 },
629
+ "mistral-large@2411-001": { inputPerMTok: 2, outputPerMTok: 6 },
630
+ "mistral-large@latest": { inputPerMTok: 2, outputPerMTok: 6 },
631
+ "mistral-medium": { inputPerMTok: 2.7, outputPerMTok: 8.1 },
632
+ "mistral-medium-2312": { inputPerMTok: 2.7, outputPerMTok: 8.1 },
633
+ "mistral-medium-2505": { inputPerMTok: 0.4, outputPerMTok: 2 },
634
+ "mistral-medium-3": { inputPerMTok: 0.4, outputPerMTok: 2 },
635
+ "mistral-medium-3-1-2508": { inputPerMTok: 0.4, outputPerMTok: 2 },
636
+ "mistral-medium-3@001": { inputPerMTok: 0.4, outputPerMTok: 2 },
637
+ "mistral-medium-latest": { inputPerMTok: 0.4, outputPerMTok: 2 },
638
+ "mistral-nemo@2407": { inputPerMTok: 3, outputPerMTok: 3 },
639
+ "mistral-nemo@latest": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
640
+ "mistral-small": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
641
+ "mistral-small-2503": { inputPerMTok: 1, outputPerMTok: 3 },
642
+ "mistral-small-2503@001": { inputPerMTok: 1, outputPerMTok: 3 },
643
+ "mistral-small-3-2-2506": { inputPerMTok: 0.06, outputPerMTok: 0.18 },
644
+ "mistral-small-latest": { inputPerMTok: 0.06, outputPerMTok: 0.18 },
645
+ "mistral-tiny": { inputPerMTok: 0.25, outputPerMTok: 0.25 },
646
+ "mistral/codestral-2405": { inputPerMTok: 1, outputPerMTok: 3 },
647
+ "mistral/codestral-2508": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
648
+ "mistral/codestral-latest": { inputPerMTok: 1, outputPerMTok: 3 },
649
+ "mistral/codestral-mamba-latest": { inputPerMTok: 0.25, outputPerMTok: 0.25 },
650
+ "mistral/devstral-2512": { inputPerMTok: 0.4, outputPerMTok: 2 },
651
+ "mistral/devstral-latest": { inputPerMTok: 0.4, outputPerMTok: 2 },
652
+ "mistral/devstral-medium-2507": { inputPerMTok: 0.4, outputPerMTok: 2 },
653
+ "mistral/devstral-medium-latest": { inputPerMTok: 0.4, outputPerMTok: 2 },
654
+ "mistral/devstral-small-2505": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
655
+ "mistral/devstral-small-2507": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
656
+ "mistral/devstral-small-latest": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
657
+ "mistral/labs-devstral-small-2512": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
658
+ "mistral/magistral-medium-1-2-2509": { inputPerMTok: 2, outputPerMTok: 5 },
659
+ "mistral/magistral-medium-2506": { inputPerMTok: 2, outputPerMTok: 5 },
660
+ "mistral/magistral-medium-2509": { inputPerMTok: 2, outputPerMTok: 5 },
661
+ "mistral/magistral-medium-latest": { inputPerMTok: 2, outputPerMTok: 5 },
662
+ "mistral/magistral-small-1-2-2509": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
663
+ "mistral/magistral-small-2506": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
664
+ "mistral/magistral-small-latest": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
665
+ "mistral/ministral-3-14b-2512": { inputPerMTok: 0.2, outputPerMTok: 0.2 },
666
+ "mistral/ministral-3-3b-2512": { inputPerMTok: 0.1, outputPerMTok: 0.1 },
667
+ "mistral/ministral-3-8b-2512": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
668
+ "mistral/ministral-8b-2512": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
669
+ "mistral/ministral-8b-latest": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
670
+ "mistral/mistral-large-2402": { inputPerMTok: 4, outputPerMTok: 12 },
671
+ "mistral/mistral-large-2407": { inputPerMTok: 3, outputPerMTok: 9 },
672
+ "mistral/mistral-large-2411": { inputPerMTok: 2, outputPerMTok: 6 },
673
+ "mistral/mistral-large-2512": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
674
+ "mistral/mistral-large-3": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
675
+ "mistral/mistral-large-latest": { inputPerMTok: 0.5, outputPerMTok: 1.5 },
676
+ "mistral/mistral-medium": { inputPerMTok: 2.7, outputPerMTok: 8.1 },
677
+ "mistral/mistral-medium-2312": { inputPerMTok: 2.7, outputPerMTok: 8.1 },
678
+ "mistral/mistral-medium-2505": { inputPerMTok: 0.4, outputPerMTok: 2 },
679
+ "mistral/mistral-medium-3-1-2508": { inputPerMTok: 0.4, outputPerMTok: 2 },
680
+ "mistral/mistral-medium-latest": { inputPerMTok: 0.4, outputPerMTok: 2 },
681
+ "mistral/mistral-small": { inputPerMTok: 0.1, outputPerMTok: 0.3 },
682
+ "mistral/mistral-small-3-2-2506": { inputPerMTok: 0.06, outputPerMTok: 0.18 },
683
+ "mistral/mistral-small-latest": { inputPerMTok: 0.06, outputPerMTok: 0.18 },
684
+ "mistral/mistral-tiny": { inputPerMTok: 0.25, outputPerMTok: 0.25 },
685
+ "mistral/open-codestral-mamba": { inputPerMTok: 0.25, outputPerMTok: 0.25 },
686
+ "mistral/open-mistral-7b": { inputPerMTok: 0.25, outputPerMTok: 0.25 },
687
+ "mistral/open-mistral-nemo": { inputPerMTok: 0.3, outputPerMTok: 0.3 },
688
+ "mistral/open-mistral-nemo-2407": { inputPerMTok: 0.3, outputPerMTok: 0.3 },
689
+ "mistral/open-mixtral-8x22b": { inputPerMTok: 2, outputPerMTok: 6 },
690
+ "mistral/open-mixtral-8x7b": { inputPerMTok: 0.7, outputPerMTok: 0.7 },
691
+ "mistral/pixtral-12b-2409": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
692
+ "mistral/pixtral-large-2411": { inputPerMTok: 2, outputPerMTok: 6 },
693
+ "mistral/pixtral-large-latest": { inputPerMTok: 2, outputPerMTok: 6 },
694
+ "mistralai/codestral-2": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
695
+ "mistralai/codestral-2@001": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
696
+ "mistralai/mistral-medium-3": { inputPerMTok: 0.4, outputPerMTok: 2 },
697
+ "mistralai/mistral-medium-3@001": { inputPerMTok: 0.4, outputPerMTok: 2 },
698
+ "o1": { inputPerMTok: 15, outputPerMTok: 60, cachedInputPerMTok: 7.5 },
699
+ "o1-2024-12-17": { inputPerMTok: 15, outputPerMTok: 60, cachedInputPerMTok: 7.5 },
700
+ "o1-pro": { inputPerMTok: 150, outputPerMTok: 600 },
701
+ "o1-pro-2025-03-19": { inputPerMTok: 150, outputPerMTok: 600 },
702
+ "o3": { inputPerMTok: 2, outputPerMTok: 8, cachedInputPerMTok: 0.5 },
703
+ "o3-2025-04-16": { inputPerMTok: 2, outputPerMTok: 8, cachedInputPerMTok: 0.5 },
704
+ "o3-deep-research": { inputPerMTok: 10, outputPerMTok: 40, cachedInputPerMTok: 2.5 },
705
+ "o3-deep-research-2025-06-26": { inputPerMTok: 10, outputPerMTok: 40, cachedInputPerMTok: 2.5 },
706
+ "o3-mini": { inputPerMTok: 1.1, outputPerMTok: 4.4, cachedInputPerMTok: 0.55 },
707
+ "o3-mini-2025-01-31": { inputPerMTok: 1.1, outputPerMTok: 4.4, cachedInputPerMTok: 0.55 },
708
+ "o3-pro": { inputPerMTok: 20, outputPerMTok: 80 },
709
+ "o3-pro-2025-06-10": { inputPerMTok: 20, outputPerMTok: 80 },
710
+ "o4-mini": { inputPerMTok: 1.1, outputPerMTok: 4.4, cachedInputPerMTok: 0.275 },
711
+ "o4-mini-2025-04-16": { inputPerMTok: 1.1, outputPerMTok: 4.4, cachedInputPerMTok: 0.275 },
712
+ "o4-mini-deep-research": { inputPerMTok: 2, outputPerMTok: 8, cachedInputPerMTok: 0.5 },
713
+ "o4-mini-deep-research-2025-06-26": { inputPerMTok: 2, outputPerMTok: 8, cachedInputPerMTok: 0.5 },
714
+ "omni-moderation-2024-09-26": { inputPerMTok: 0, outputPerMTok: 0 },
715
+ "omni-moderation-latest": { inputPerMTok: 0, outputPerMTok: 0 },
716
+ "open-codestral-mamba": { inputPerMTok: 0.25, outputPerMTok: 0.25 },
717
+ "open-mistral-7b": { inputPerMTok: 0.25, outputPerMTok: 0.25 },
718
+ "open-mistral-nemo": { inputPerMTok: 0.3, outputPerMTok: 0.3 },
719
+ "open-mistral-nemo-2407": { inputPerMTok: 0.3, outputPerMTok: 0.3 },
720
+ "open-mixtral-8x22b": { inputPerMTok: 2, outputPerMTok: 6 },
721
+ "open-mixtral-8x7b": { inputPerMTok: 0.7, outputPerMTok: 0.7 },
722
+ "pixtral-12b-2409": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
723
+ "pixtral-large-2411": { inputPerMTok: 2, outputPerMTok: 6 },
724
+ "pixtral-large-latest": { inputPerMTok: 2, outputPerMTok: 6 },
725
+ "rerank-english-v2.0": { inputPerMTok: 0, outputPerMTok: 0 },
726
+ "rerank-english-v3.0": { inputPerMTok: 0, outputPerMTok: 0 },
727
+ "rerank-multilingual-v2.0": { inputPerMTok: 0, outputPerMTok: 0 },
728
+ "rerank-multilingual-v3.0": { inputPerMTok: 0, outputPerMTok: 0 },
729
+ "rerank-v3.5": { inputPerMTok: 0, outputPerMTok: 0 },
730
+ "text-embedding-3-large": { inputPerMTok: 0.13, outputPerMTok: 0 },
731
+ "text-embedding-3-small": { inputPerMTok: 0.02, outputPerMTok: 0 },
732
+ "text-embedding-ada-002": { inputPerMTok: 0.1, outputPerMTok: 0 },
733
+ "text-embedding-ada-002-v2": { inputPerMTok: 0.1, outputPerMTok: 0 },
734
+ "text-moderation-007": { inputPerMTok: 0, outputPerMTok: 0 },
735
+ "text-moderation-latest": { inputPerMTok: 0, outputPerMTok: 0 },
736
+ "text-moderation-stable": { inputPerMTok: 0, outputPerMTok: 0 },
737
+ "vertex_ai/claude-3-5-haiku": { inputPerMTok: 1, outputPerMTok: 5 },
738
+ "vertex_ai/claude-3-5-haiku@20241022": { inputPerMTok: 1, outputPerMTok: 5 },
739
+ "vertex_ai/claude-3-5-sonnet": { inputPerMTok: 3, outputPerMTok: 15 },
740
+ "vertex_ai/claude-3-5-sonnet@20240620": { inputPerMTok: 3, outputPerMTok: 15 },
741
+ "vertex_ai/claude-3-7-sonnet@20250219": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
742
+ "vertex_ai/claude-3-haiku": { inputPerMTok: 0.25, outputPerMTok: 1.25 },
743
+ "vertex_ai/claude-3-haiku@20240307": { inputPerMTok: 0.25, outputPerMTok: 1.25 },
744
+ "vertex_ai/claude-3-opus": { inputPerMTok: 15, outputPerMTok: 75 },
745
+ "vertex_ai/claude-3-opus@20240229": { inputPerMTok: 15, outputPerMTok: 75 },
746
+ "vertex_ai/claude-3-sonnet": { inputPerMTok: 3, outputPerMTok: 15 },
747
+ "vertex_ai/claude-3-sonnet@20240229": { inputPerMTok: 3, outputPerMTok: 15 },
748
+ "vertex_ai/claude-haiku-4-5": { inputPerMTok: 1, outputPerMTok: 5, cachedInputPerMTok: 0.1 },
749
+ "vertex_ai/claude-haiku-4-5@20251001": { inputPerMTok: 1, outputPerMTok: 5, cachedInputPerMTok: 0.1 },
750
+ "vertex_ai/claude-opus-4": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
751
+ "vertex_ai/claude-opus-4-1": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
752
+ "vertex_ai/claude-opus-4-1@20250805": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
753
+ "vertex_ai/claude-opus-4-5": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
754
+ "vertex_ai/claude-opus-4-5@20251101": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
755
+ "vertex_ai/claude-opus-4-6": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
756
+ "vertex_ai/claude-opus-4-6@default": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
757
+ "vertex_ai/claude-opus-4-7": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
758
+ "vertex_ai/claude-opus-4-7@default": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
759
+ "vertex_ai/claude-opus-4-8": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
760
+ "vertex_ai/claude-opus-4-8@default": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 0.5 },
761
+ "vertex_ai/claude-opus-4@20250514": { inputPerMTok: 15, outputPerMTok: 75, cachedInputPerMTok: 1.5 },
762
+ "vertex_ai/claude-sonnet-4": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
763
+ "vertex_ai/claude-sonnet-4-5": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
764
+ "vertex_ai/claude-sonnet-4-5@20250929": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
765
+ "vertex_ai/claude-sonnet-4-6": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
766
+ "vertex_ai/claude-sonnet-4-6@default": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
767
+ "vertex_ai/claude-sonnet-4@20250514": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.3 },
768
+ "vertex_ai/codestral-2": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
769
+ "vertex_ai/codestral-2501": { inputPerMTok: 0.2, outputPerMTok: 0.6 },
770
+ "vertex_ai/codestral-2@001": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
771
+ "vertex_ai/codestral@2405": { inputPerMTok: 0.2, outputPerMTok: 0.6 },
772
+ "vertex_ai/codestral@latest": { inputPerMTok: 0.2, outputPerMTok: 0.6 },
773
+ "vertex_ai/deep-research-pro-preview-12-2025": { inputPerMTok: 2, outputPerMTok: 12 },
774
+ "vertex_ai/gemini-2.5-flash-image": { inputPerMTok: 0.3, outputPerMTok: 2.5, cachedInputPerMTok: 0.03 },
775
+ "vertex_ai/gemini-3-pro-image-preview": { inputPerMTok: 2, outputPerMTok: 12 },
776
+ "vertex_ai/gemini-3.1-flash-image-preview": { inputPerMTok: 0.5, outputPerMTok: 3 },
777
+ "vertex_ai/gemini-3.1-flash-lite": { inputPerMTok: 0.25, outputPerMTok: 1.5, cachedInputPerMTok: 0.025 },
778
+ "vertex_ai/gemini-3.1-flash-lite-preview": { inputPerMTok: 0.25, outputPerMTok: 1.5, cachedInputPerMTok: 0.025 },
779
+ "vertex_ai/mistral-large-2411": { inputPerMTok: 2, outputPerMTok: 6 },
780
+ "vertex_ai/mistral-large@2407": { inputPerMTok: 2, outputPerMTok: 6 },
781
+ "vertex_ai/mistral-large@2411-001": { inputPerMTok: 2, outputPerMTok: 6 },
782
+ "vertex_ai/mistral-large@latest": { inputPerMTok: 2, outputPerMTok: 6 },
783
+ "vertex_ai/mistral-medium-3": { inputPerMTok: 0.4, outputPerMTok: 2 },
784
+ "vertex_ai/mistral-medium-3@001": { inputPerMTok: 0.4, outputPerMTok: 2 },
785
+ "vertex_ai/mistral-nemo@2407": { inputPerMTok: 3, outputPerMTok: 3 },
786
+ "vertex_ai/mistral-nemo@latest": { inputPerMTok: 0.15, outputPerMTok: 0.15 },
787
+ "vertex_ai/mistral-small-2503": { inputPerMTok: 1, outputPerMTok: 3 },
788
+ "vertex_ai/mistral-small-2503@001": { inputPerMTok: 1, outputPerMTok: 3 },
789
+ "vertex_ai/mistralai/codestral-2": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
790
+ "vertex_ai/mistralai/codestral-2@001": { inputPerMTok: 0.3, outputPerMTok: 0.9 },
791
+ "vertex_ai/mistralai/mistral-medium-3": { inputPerMTok: 0.4, outputPerMTok: 2 },
792
+ "vertex_ai/mistralai/mistral-medium-3@001": { inputPerMTok: 0.4, outputPerMTok: 2 },
793
+ "xai/grok-2": { inputPerMTok: 2, outputPerMTok: 10 },
794
+ "xai/grok-2-1212": { inputPerMTok: 2, outputPerMTok: 10 },
795
+ "xai/grok-2-latest": { inputPerMTok: 2, outputPerMTok: 10 },
796
+ "xai/grok-2-vision": { inputPerMTok: 2, outputPerMTok: 10 },
797
+ "xai/grok-2-vision-1212": { inputPerMTok: 2, outputPerMTok: 10 },
798
+ "xai/grok-2-vision-latest": { inputPerMTok: 2, outputPerMTok: 10 },
799
+ "xai/grok-3": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.75 },
800
+ "xai/grok-3-beta": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.75 },
801
+ "xai/grok-3-fast-beta": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 1.25 },
802
+ "xai/grok-3-fast-latest": { inputPerMTok: 5, outputPerMTok: 25, cachedInputPerMTok: 1.25 },
803
+ "xai/grok-3-latest": { inputPerMTok: 3, outputPerMTok: 15, cachedInputPerMTok: 0.75 },
804
+ "xai/grok-3-mini": { inputPerMTok: 0.3, outputPerMTok: 0.5, cachedInputPerMTok: 0.075 },
805
+ "xai/grok-3-mini-beta": { inputPerMTok: 0.3, outputPerMTok: 0.5, cachedInputPerMTok: 0.075 },
806
+ "xai/grok-3-mini-fast": { inputPerMTok: 0.6, outputPerMTok: 4, cachedInputPerMTok: 0.15 },
807
+ "xai/grok-3-mini-fast-beta": { inputPerMTok: 0.6, outputPerMTok: 4, cachedInputPerMTok: 0.15 },
808
+ "xai/grok-3-mini-fast-latest": { inputPerMTok: 0.6, outputPerMTok: 4, cachedInputPerMTok: 0.15 },
809
+ "xai/grok-3-mini-latest": { inputPerMTok: 0.3, outputPerMTok: 0.5, cachedInputPerMTok: 0.075 },
810
+ "xai/grok-4": { inputPerMTok: 3, outputPerMTok: 15 },
811
+ "xai/grok-4-0709": { inputPerMTok: 3, outputPerMTok: 15 },
812
+ "xai/grok-4-1-fast": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
813
+ "xai/grok-4-1-fast-non-reasoning": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
814
+ "xai/grok-4-1-fast-non-reasoning-latest": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
815
+ "xai/grok-4-1-fast-reasoning": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
816
+ "xai/grok-4-1-fast-reasoning-latest": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
817
+ "xai/grok-4-fast-non-reasoning": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
818
+ "xai/grok-4-fast-reasoning": { inputPerMTok: 0.2, outputPerMTok: 0.5, cachedInputPerMTok: 0.05 },
819
+ "xai/grok-4-latest": { inputPerMTok: 3, outputPerMTok: 15 },
820
+ "xai/grok-4.20-0309-reasoning": { inputPerMTok: 2, outputPerMTok: 6, cachedInputPerMTok: 0.2 },
821
+ "xai/grok-4.20-beta-0309-non-reasoning": { inputPerMTok: 2, outputPerMTok: 6, cachedInputPerMTok: 0.2 },
822
+ "xai/grok-4.20-beta-0309-reasoning": { inputPerMTok: 2, outputPerMTok: 6, cachedInputPerMTok: 0.2 },
823
+ "xai/grok-4.20-multi-agent-beta-0309": { inputPerMTok: 2, outputPerMTok: 6, cachedInputPerMTok: 0.2 },
824
+ "xai/grok-4.3": { inputPerMTok: 1.25, outputPerMTok: 2.5, cachedInputPerMTok: 0.2 },
825
+ "xai/grok-4.3-latest": { inputPerMTok: 1.25, outputPerMTok: 2.5, cachedInputPerMTok: 0.2 },
826
+ "xai/grok-beta": { inputPerMTok: 5, outputPerMTok: 15 },
827
+ "xai/grok-code-fast": { inputPerMTok: 0.2, outputPerMTok: 1.5, cachedInputPerMTok: 0.02 },
828
+ "xai/grok-code-fast-1": { inputPerMTok: 0.2, outputPerMTok: 1.5, cachedInputPerMTok: 0.02 },
829
+ "xai/grok-code-fast-1-0825": { inputPerMTok: 0.2, outputPerMTok: 1.5, cachedInputPerMTok: 0.02 },
830
+ "xai/grok-vision-beta": { inputPerMTok: 5, outputPerMTok: 15 }
831
+ };
832
+
833
+ // src/fetch-prices.ts
834
+ var LITELLM_URL = "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json";
835
+ var TARGET_PROVIDERS = /* @__PURE__ */ new Set([
836
+ "anthropic",
837
+ "openai",
838
+ "gemini",
839
+ "deepseek",
840
+ "mistral",
841
+ "xai",
842
+ "cohere",
843
+ "vertex_ai-anthropic_models",
844
+ "vertex_ai-language-models",
845
+ "vertex_ai-mistral_models"
846
+ ]);
847
+ var STRIP_PREFIXES = [
848
+ "gemini/",
849
+ "vertex_ai/",
850
+ "anthropic/",
851
+ "openai/",
852
+ "xai/",
853
+ "cohere/",
854
+ "mistral/",
855
+ "deepseek/"
856
+ ];
857
+ function roundPrice(v) {
858
+ if (v === 0) return 0;
859
+ const mag = Math.floor(Math.log10(Math.abs(v)));
860
+ const factor = Math.pow(10, 5 - mag);
861
+ return Math.round(v * factor) / factor;
862
+ }
863
+ function mapLiteLLM(raw) {
864
+ const out = {};
865
+ for (const [key, val] of Object.entries(raw)) {
866
+ if (!val || typeof val !== "object") continue;
867
+ const entry = val;
868
+ const prov = entry.litellm_provider ?? "";
869
+ if (!TARGET_PROVIDERS.has(prov)) continue;
870
+ const inputCost = entry.input_cost_per_token;
871
+ const outputCost = entry.output_cost_per_token;
872
+ if (inputCost == null || outputCost == null) continue;
873
+ const priceEntry = {
874
+ inputPerMTok: roundPrice(inputCost * 1e6),
875
+ outputPerMTok: roundPrice(outputCost * 1e6)
876
+ };
877
+ if (entry.cache_read_input_token_cost != null) {
878
+ priceEntry.cachedInputPerMTok = roundPrice(entry.cache_read_input_token_cost * 1e6);
879
+ }
880
+ out[key] = priceEntry;
881
+ for (const prefix of STRIP_PREFIXES) {
882
+ if (key.startsWith(prefix)) {
883
+ const bare = key.slice(prefix.length);
884
+ if (!(bare in out)) out[bare] = priceEntry;
885
+ break;
886
+ }
887
+ }
888
+ }
889
+ return out;
890
+ }
891
+ async function fetchLatestPrices(opts) {
892
+ const fetchFn = opts?.fetchImpl ?? fetch;
893
+ const url = opts?.url ?? LITELLM_URL;
894
+ let res;
895
+ try {
896
+ res = await fetchFn(url);
897
+ } catch (e) {
898
+ throw new Error(
899
+ `fetchLatestPrices: network request to ${url} failed \u2014 ${e instanceof Error ? e.message : String(e)}`
900
+ );
901
+ }
902
+ if (!res.ok) {
903
+ throw new Error(`fetchLatestPrices: HTTP ${res.status} ${res.statusText} from ${url}`);
904
+ }
905
+ let raw;
906
+ try {
907
+ raw = await res.json();
908
+ } catch (e) {
909
+ throw new Error(
910
+ `fetchLatestPrices: failed to parse JSON from ${url} \u2014 ${e instanceof Error ? e.message : String(e)}`
911
+ );
912
+ }
913
+ return mapLiteLLM(raw);
914
+ }
915
+
916
+ // src/ollama.ts
917
+ var import_node_module = require("node:module");
918
+ var import_meta = {};
919
+ function loadOllamaModule() {
920
+ const req = typeof require === "function" ? require : (0, import_node_module.createRequire)(import_meta.url);
921
+ try {
922
+ return req("ollama-ai-provider");
923
+ } catch {
924
+ throw new Error(
925
+ "[eidentic/model] createOllamaModel requires the `ollama-ai-provider` package.\nInstall it in your project:\n npm install ollama-ai-provider\n # or\n pnpm add ollama-ai-provider"
926
+ );
927
+ }
928
+ }
929
+ function createOllamaModel(modelId, opts) {
930
+ const mod = opts?._factory ?? loadOllamaModule();
931
+ const provider = mod.createOllama(opts?.baseURL ? { baseURL: opts.baseURL } : void 0);
932
+ return provider(modelId);
933
+ }
934
+
935
+ // src/fallback.ts
936
+ function isAbortError(err) {
937
+ if (err instanceof DOMException && err.name === "AbortError") return true;
938
+ if (err instanceof Error && err.name === "AbortError") return true;
939
+ return false;
940
+ }
941
+ function shouldTryNext(err, opts) {
942
+ if (isAbortError(err)) return false;
943
+ if (opts?.shouldFallback) return opts.shouldFallback(err);
944
+ return true;
945
+ }
946
+ function withFallback(primary, fallbacks, opts) {
947
+ const chain = [primary, ...fallbacks];
948
+ return {
949
+ // Expose the primary's modelId so AgentConfig can read it.
950
+ get modelId() {
951
+ return primary.modelId;
952
+ },
953
+ async complete(request) {
954
+ let lastErr;
955
+ for (let i = 0; i < chain.length; i++) {
956
+ try {
957
+ return await chain[i].complete(request);
958
+ } catch (err) {
959
+ lastErr = err;
960
+ if (i < chain.length - 1 && shouldTryNext(err, opts)) {
961
+ opts?.onFallback?.(err, i, i + 1);
962
+ continue;
963
+ }
964
+ throw err;
965
+ }
966
+ }
967
+ throw lastErr;
968
+ },
969
+ async *stream(request) {
970
+ let lastErr;
971
+ for (let i = 0; i < chain.length; i++) {
972
+ const model = chain[i];
973
+ const streamFn = model.stream?.bind(model);
974
+ if (!streamFn) {
975
+ lastErr = new Error(`model at index ${i} does not implement stream()`);
976
+ if (i < chain.length - 1 && shouldTryNext(lastErr, opts)) {
977
+ opts?.onFallback?.(lastErr, i, i + 1);
978
+ continue;
979
+ }
980
+ throw lastErr;
981
+ }
982
+ let deltasSeen = 0;
983
+ try {
984
+ for await (const part of streamFn(request)) {
985
+ if (part.type === "delta") deltasSeen++;
986
+ yield part;
987
+ }
988
+ return;
989
+ } catch (err) {
990
+ lastErr = err;
991
+ if (deltasSeen > 0) throw err;
992
+ if (i < chain.length - 1 && shouldTryNext(err, opts)) {
993
+ opts?.onFallback?.(err, i, i + 1);
994
+ continue;
995
+ }
996
+ throw err;
997
+ }
998
+ }
999
+ throw lastErr;
1000
+ }
1001
+ };
1002
+ }
1003
+
1004
+ // src/route.ts
1005
+ function estimateTokens(req) {
1006
+ let chars = 0;
1007
+ for (const msg of req.messages) {
1008
+ if (typeof msg.content === "string") {
1009
+ chars += msg.content.length;
1010
+ } else {
1011
+ for (const block of msg.content) {
1012
+ if ("text" in block && typeof block.text === "string") chars += block.text.length;
1013
+ }
1014
+ }
1015
+ }
1016
+ return Math.ceil(chars / 4);
1017
+ }
1018
+ function byTokenEstimate(thresholds, fallbackTier) {
1019
+ const sorted = [...thresholds].sort((a, b) => a.upTo - b.upTo);
1020
+ return (req) => {
1021
+ const tokens = estimateTokens(req);
1022
+ for (const t of sorted) {
1023
+ if (tokens <= t.upTo) return t.tier;
1024
+ }
1025
+ return fallbackTier;
1026
+ };
1027
+ }
1028
+ function routeModel(selector, tiers) {
1029
+ function resolve(req) {
1030
+ const tier = selector(req);
1031
+ const model = tiers[tier];
1032
+ if (!model) {
1033
+ const known = Object.keys(tiers).join(", ");
1034
+ throw new Error(
1035
+ `routeModel: unknown tier "${tier}". Known tiers: ${known || "(none)"}`
1036
+ );
1037
+ }
1038
+ return model;
1039
+ }
1040
+ return {
1041
+ get modelId() {
1042
+ return void 0;
1043
+ },
1044
+ async complete(request) {
1045
+ return resolve(request).complete(request);
1046
+ },
1047
+ async *stream(request) {
1048
+ const model = resolve(request);
1049
+ if (!model.stream) {
1050
+ throw new Error(
1051
+ `routeModel: resolved model for tier does not implement stream()`
1052
+ );
1053
+ }
1054
+ yield* model.stream(request);
1055
+ }
1056
+ };
1057
+ }
1058
+
1059
+ // src/cache.ts
1060
+ function stableJson(value) {
1061
+ if (value === null || typeof value !== "object") return JSON.stringify(value);
1062
+ if (Array.isArray(value)) return `[${value.map(stableJson).join(",")}]`;
1063
+ const sorted = Object.keys(value).sort();
1064
+ return `{${sorted.map((k) => `${JSON.stringify(k)}:${stableJson(value[k])}`).join(",")}}`;
1065
+ }
1066
+ function defaultKeyFn(req) {
1067
+ return stableJson({
1068
+ messages: req.messages,
1069
+ tools: req.tools,
1070
+ model: req.model ?? null,
1071
+ outputSchema: req.outputSchema ?? null
1072
+ });
1073
+ }
1074
+ var LRUCache = class {
1075
+ constructor(maxEntries) {
1076
+ this.maxEntries = maxEntries;
1077
+ }
1078
+ maxEntries;
1079
+ map = /* @__PURE__ */ new Map();
1080
+ get(key) {
1081
+ const entry = this.map.get(key);
1082
+ if (!entry) return void 0;
1083
+ this.map.delete(key);
1084
+ this.map.set(key, entry);
1085
+ return entry;
1086
+ }
1087
+ set(key, entry) {
1088
+ if (this.map.has(key)) this.map.delete(key);
1089
+ this.map.set(key, entry);
1090
+ if (this.map.size > this.maxEntries) {
1091
+ const oldest = this.map.keys().next().value;
1092
+ if (oldest !== void 0) this.map.delete(oldest);
1093
+ }
1094
+ }
1095
+ get size() {
1096
+ return this.map.size;
1097
+ }
1098
+ };
1099
+ function cachedModel(model, opts = {}) {
1100
+ const ttlMs = opts.ttlMs ?? 5 * 6e4;
1101
+ const maxEntries = opts.maxEntries ?? 500;
1102
+ const keyFn = opts.keyFn ?? defaultKeyFn;
1103
+ const shouldCache = opts.shouldCache ?? (() => true);
1104
+ const clock = opts.clock ?? Date.now;
1105
+ const lru = new LRUCache(maxEntries);
1106
+ let hits = 0;
1107
+ let misses = 0;
1108
+ async function getFromCache(key) {
1109
+ const now = clock();
1110
+ const local = lru.get(key);
1111
+ if (local) {
1112
+ if (local.expiresAt > now) return local.value;
1113
+ }
1114
+ if (opts.store) {
1115
+ const stored = await opts.store.get(key);
1116
+ if (stored !== void 0) {
1117
+ lru.set(key, { value: stored, expiresAt: now + ttlMs });
1118
+ return stored;
1119
+ }
1120
+ }
1121
+ return void 0;
1122
+ }
1123
+ async function setInCache(key, value) {
1124
+ const expiresAt = ttlMs === Infinity ? Infinity : clock() + ttlMs;
1125
+ lru.set(key, { value, expiresAt });
1126
+ if (opts.store) {
1127
+ await opts.store.set(key, value, ttlMs === Infinity ? void 0 : ttlMs);
1128
+ }
1129
+ }
1130
+ const port = {
1131
+ get modelId() {
1132
+ return model.modelId;
1133
+ },
1134
+ async complete(request) {
1135
+ const key = keyFn(request);
1136
+ const cached = await getFromCache(key);
1137
+ if (cached !== void 0) {
1138
+ hits++;
1139
+ opts.onCacheHit?.(key, cached);
1140
+ return cached;
1141
+ }
1142
+ misses++;
1143
+ const result = await model.complete(request);
1144
+ if (shouldCache(request, result)) {
1145
+ await setInCache(key, result);
1146
+ }
1147
+ return result;
1148
+ },
1149
+ async *stream(request) {
1150
+ if (!model.stream) {
1151
+ throw new Error("cachedModel: wrapped model does not implement stream()");
1152
+ }
1153
+ yield* model.stream(request);
1154
+ },
1155
+ stats() {
1156
+ return { hits, misses, size: lru.size };
1157
+ }
1158
+ };
1159
+ return port;
1160
+ }
1161
+ // Annotate the CommonJS export names for ESM import in node:
1162
+ 0 && (module.exports = {
1163
+ AIEmbedder,
1164
+ AIModel,
1165
+ LITELLM_URL,
1166
+ byTokenEstimate,
1167
+ cachedModel,
1168
+ createOllamaModel,
1169
+ defaultPrices,
1170
+ fetchLatestPrices,
1171
+ mapLiteLLM,
1172
+ pricesUpdatedAt,
1173
+ routeModel,
1174
+ withFallback
1175
+ });