llm.rb 15.0.3 → 15.2.0

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 (104) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGELOG.md +433 -3
  3. data/README.md +188 -71
  4. data/bin/llm.rb +50 -7
  5. data/data/alibaba.json +912 -823
  6. data/data/anthropic.json +234 -187
  7. data/data/bedrock.json +3702 -2058
  8. data/data/deepinfra.json +1288 -951
  9. data/data/deepseek.json +87 -53
  10. data/data/google.json +670 -670
  11. data/data/mistral.json +501 -460
  12. data/data/moonshot.json +43 -248
  13. data/data/openai.json +1008 -914
  14. data/data/openrouter.json +14417 -0
  15. data/data/xai.json +213 -201
  16. data/data/zai.json +242 -149
  17. data/docs/deepdive/advanced/compaction.md +5 -5
  18. data/docs/deepdive/advanced/context.md +8 -6
  19. data/docs/deepdive/advanced/guard.md +2 -2
  20. data/docs/deepdive/features/builtin_tools.md +93 -22
  21. data/docs/deepdive/features/{repl.md → console.md} +28 -28
  22. data/docs/deepdive/features/database.md +3 -3
  23. data/docs/deepdive/fundamentals/agents.md +13 -12
  24. data/docs/deepdive/fundamentals/providers.md +91 -6
  25. data/docs/deepdive/fundamentals/skills.md +14 -6
  26. data/docs/deepdive/fundamentals/stream.md +4 -4
  27. data/docs/deepdive/fundamentals/tools.md +63 -31
  28. data/docs/deepdive/reference/cost.md +2 -2
  29. data/docs/deepdive/reference/model_registry.md +2 -2
  30. data/docs/deepdive/reference/tracer.md +15 -13
  31. data/docs/deepdive.md +2 -2
  32. data/lib/llm/active_record/acts_as_agent.rb +9 -5
  33. data/lib/llm/agent.rb +40 -15
  34. data/lib/llm/{repl → console}/bar.rb +3 -3
  35. data/lib/llm/{repl → console}/buffer.rb +24 -9
  36. data/lib/llm/{repl → console}/color.rb +2 -2
  37. data/lib/llm/{repl → console}/command.rb +12 -12
  38. data/lib/llm/{repl → console}/commands/exit.rb +4 -4
  39. data/lib/llm/{repl → console}/commands/help.rb +1 -1
  40. data/lib/llm/{repl/commands/compact.rb → console/commands/keep.rb} +11 -9
  41. data/lib/llm/{repl → console}/commands/model.rb +2 -2
  42. data/lib/llm/{repl → console}/input/cache.rb +2 -2
  43. data/lib/llm/{repl → console}/input/char.rb +2 -2
  44. data/lib/llm/{repl → console}/input/row.rb +1 -1
  45. data/lib/llm/{repl → console}/input.rb +18 -10
  46. data/lib/llm/console/markdown/parser.rb +78 -0
  47. data/lib/llm/{repl → console}/markdown/table.rb +8 -5
  48. data/lib/llm/{repl → console}/markdown.rb +13 -30
  49. data/lib/llm/console/node.rb +69 -0
  50. data/lib/llm/{repl → console}/status.rb +11 -11
  51. data/lib/llm/{repl → console}/stream.rb +36 -9
  52. data/lib/llm/{repl → console}/walker.rb +1 -1
  53. data/lib/llm/{repl → console}/window.rb +17 -17
  54. data/lib/llm/{repl.rb → console.rb} +39 -19
  55. data/lib/llm/context/deserializer.rb +2 -1
  56. data/lib/llm/context.rb +29 -13
  57. data/lib/llm/cost.rb +13 -0
  58. data/lib/llm/function/async/reactor.rb +20 -1
  59. data/lib/llm/function/fork/task.rb +14 -10
  60. data/lib/llm/function.rb +1 -1
  61. data/lib/llm/json_adapter.rb +40 -28
  62. data/lib/llm/message.rb +7 -0
  63. data/lib/llm/provider.rb +31 -10
  64. data/lib/llm/providers/alibaba.rb +1 -1
  65. data/lib/llm/providers/anthropic.rb +1 -1
  66. data/lib/llm/providers/bedrock/models.rb +2 -2
  67. data/lib/llm/providers/bedrock.rb +1 -1
  68. data/lib/llm/providers/deepseek.rb +1 -1
  69. data/lib/llm/providers/google.rb +1 -1
  70. data/lib/llm/providers/ollama.rb +1 -1
  71. data/lib/llm/providers/openai/responses.rb +2 -1
  72. data/lib/llm/providers/openai.rb +2 -1
  73. data/lib/llm/providers/openrouter.rb +87 -0
  74. data/lib/llm/schema/leaf.rb +34 -2
  75. data/lib/llm/schema.rb +4 -2
  76. data/lib/llm/sequel/agent.rb +9 -5
  77. data/lib/llm/skill.rb +7 -1
  78. data/lib/llm/stream.rb +8 -3
  79. data/lib/llm/tool/param.rb +5 -1
  80. data/lib/llm/tool.rb +5 -0
  81. data/lib/llm/tools/bundle.rb +53 -0
  82. data/lib/llm/tools/edit-file.rb +7 -2
  83. data/lib/llm/tools/exec.rb +78 -0
  84. data/lib/llm/tools/git.rb +27 -26
  85. data/lib/llm/tools/mkdir.rb +12 -19
  86. data/lib/llm/tools/read_file.rb +69 -9
  87. data/lib/llm/tools/rg.rb +20 -24
  88. data/lib/llm/tools/ruby.rb +17 -25
  89. data/lib/llm/tools/utils.rb +75 -2
  90. data/lib/llm/tools/write_file.rb +4 -1
  91. data/lib/llm/tracer/logger.rb +2 -2
  92. data/lib/llm/tracer/pretty_logger.rb +4 -4
  93. data/lib/llm/tracer/telemetry.rb +2 -2
  94. data/lib/llm/tracer.rb +33 -0
  95. data/lib/llm/transport/curb.rb +5 -3
  96. data/lib/llm/transport/http.rb +5 -2
  97. data/lib/llm/transport/persistent_http.rb +6 -4
  98. data/lib/llm/transport/utils.rb +8 -6
  99. data/lib/llm/version.rb +1 -1
  100. data/lib/llm.rb +18 -12
  101. data/llm.gemspec +8 -8
  102. metadata +80 -37
  103. data/lib/llm/repl/node.rb +0 -44
  104. data/lib/llm/tools/shell.rb +0 -55
data/data/mistral.json CHANGED
@@ -7,58 +7,18 @@
7
7
  "name": "Mistral",
8
8
  "doc": "https://docs.mistral.ai/getting-started/models/",
9
9
  "models": {
10
- "mistral-medium-2508": {
11
- "id": "mistral-medium-2508",
12
- "name": "Mistral Medium 3.1",
13
- "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
14
- "family": "mistral-medium",
10
+ "pixtral-12b": {
11
+ "id": "pixtral-12b",
12
+ "name": "Pixtral 12B",
13
+ "description": "Mistral vision-language model for image understanding and multimodal chat",
14
+ "family": "pixtral",
15
15
  "attachment": true,
16
16
  "reasoning": false,
17
17
  "tool_call": true,
18
18
  "temperature": true,
19
- "knowledge": "2025-05",
20
- "release_date": "2025-08-12",
21
- "last_updated": "2025-08-12",
22
- "modalities": {
23
- "input": [
24
- "text",
25
- "image"
26
- ],
27
- "output": [
28
- "text"
29
- ]
30
- },
31
- "open_weights": false,
32
- "limit": {
33
- "context": 262144,
34
- "output": 262144
35
- },
36
- "cost": {
37
- "input": 0.4,
38
- "output": 2
39
- }
40
- },
41
- "mistral-small-latest": {
42
- "id": "mistral-small-latest",
43
- "name": "Mistral Small (latest)",
44
- "description": "Efficient Mistral model for fast chat, extraction, and production assistants",
45
- "family": "mistral-small",
46
- "attachment": true,
47
- "reasoning": true,
48
- "reasoning_options": [
49
- {
50
- "type": "effort",
51
- "values": [
52
- "none",
53
- "high"
54
- ]
55
- }
56
- ],
57
- "tool_call": true,
58
- "temperature": true,
59
- "knowledge": "2025-06",
60
- "release_date": "2026-03-16",
61
- "last_updated": "2026-03-16",
19
+ "knowledge": "2024-09",
20
+ "release_date": "2024-09-01",
21
+ "last_updated": "2024-09-01",
62
22
  "modalities": {
63
23
  "input": [
64
24
  "text",
@@ -70,17 +30,17 @@
70
30
  },
71
31
  "open_weights": true,
72
32
  "limit": {
73
- "context": 256000,
74
- "output": 256000
33
+ "context": 128000,
34
+ "output": 128000
75
35
  },
76
36
  "cost": {
77
37
  "input": 0.15,
78
- "output": 0.6
38
+ "output": 0.15
79
39
  }
80
40
  },
81
- "devstral-medium-2507": {
82
- "id": "devstral-medium-2507",
83
- "name": "Devstral Medium",
41
+ "devstral-small-2507": {
42
+ "id": "devstral-small-2507",
43
+ "name": "Devstral Small",
84
44
  "description": "Legacy model retained for compatibility with older integrations",
85
45
  "family": "devstral",
86
46
  "attachment": false,
@@ -105,55 +65,26 @@
105
65
  },
106
66
  "status": "deprecated",
107
67
  "cost": {
108
- "input": 0.4,
109
- "output": 2
110
- }
111
- },
112
- "open-mixtral-8x7b": {
113
- "id": "open-mixtral-8x7b",
114
- "name": "Mixtral 8x7B",
115
- "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
116
- "family": "mixtral",
117
- "attachment": false,
118
- "reasoning": false,
119
- "tool_call": true,
120
- "temperature": true,
121
- "knowledge": "2024-01",
122
- "release_date": "2023-12-11",
123
- "last_updated": "2023-12-11",
124
- "modalities": {
125
- "input": [
126
- "text"
127
- ],
128
- "output": [
129
- "text"
130
- ]
131
- },
132
- "open_weights": true,
133
- "limit": {
134
- "context": 32000,
135
- "output": 32000
136
- },
137
- "cost": {
138
- "input": 0.7,
139
- "output": 0.7
68
+ "input": 0.1,
69
+ "output": 0.3
140
70
  }
141
71
  },
142
- "mistral-large-2411": {
143
- "id": "mistral-large-2411",
144
- "name": "Mistral Large 2.1",
145
- "description": "Flagship Mistral model for advanced reasoning, coding, and multilingual work",
146
- "family": "mistral-large",
72
+ "mistral-small-2506": {
73
+ "id": "mistral-small-2506",
74
+ "name": "Mistral Small 3.2",
75
+ "description": "Efficient Mistral model for fast chat, extraction, and production assistants",
76
+ "family": "mistral-small",
147
77
  "attachment": false,
148
78
  "reasoning": false,
149
79
  "tool_call": true,
150
80
  "temperature": true,
151
- "knowledge": "2024-11",
152
- "release_date": "2024-11-18",
153
- "last_updated": "2024-11-18",
81
+ "knowledge": "2025-03",
82
+ "release_date": "2025-06-20",
83
+ "last_updated": "2025-06-20",
154
84
  "modalities": {
155
85
  "input": [
156
- "text"
86
+ "text",
87
+ "image"
157
88
  ],
158
89
  "output": [
159
90
  "text"
@@ -161,26 +92,27 @@
161
92
  },
162
93
  "open_weights": true,
163
94
  "limit": {
164
- "context": 131072,
95
+ "context": 128000,
165
96
  "output": 16384
166
97
  },
167
98
  "cost": {
168
- "input": 2,
169
- "output": 6
99
+ "input": 0.1,
100
+ "output": 0.3
170
101
  }
171
102
  },
172
- "mistral-nemo": {
173
- "id": "mistral-nemo",
174
- "name": "Mistral Nemo",
175
- "description": "Efficient Mistral-NVIDIA open model for multilingual chat and local deployment",
176
- "family": "mistral-nemo",
103
+ "magistral-small": {
104
+ "id": "magistral-small",
105
+ "name": "Magistral Small",
106
+ "description": "Mistral reasoning model for transparent analysis, math, and complex decisions",
107
+ "family": "magistral-small",
177
108
  "attachment": false,
178
- "reasoning": false,
109
+ "reasoning": true,
110
+ "reasoning_options": [],
179
111
  "tool_call": true,
180
112
  "temperature": true,
181
- "knowledge": "2024-07",
182
- "release_date": "2024-07-01",
183
- "last_updated": "2024-07-01",
113
+ "knowledge": "2025-06",
114
+ "release_date": "2025-03-17",
115
+ "last_updated": "2025-03-17",
184
116
  "modalities": {
185
117
  "input": [
186
118
  "text"
@@ -195,14 +127,14 @@
195
127
  "output": 128000
196
128
  },
197
129
  "cost": {
198
- "input": 0.15,
199
- "output": 0.15
130
+ "input": 0.5,
131
+ "output": 1.5
200
132
  }
201
133
  },
202
- "devstral-latest": {
203
- "id": "devstral-latest",
134
+ "devstral-2512": {
135
+ "id": "devstral-2512",
204
136
  "name": "Devstral 2",
205
- "description": "Legacy model retained for compatibility with older integrations",
137
+ "description": "Mistral's coding-agent model for repository work, terminal tasks, and software fixes",
206
138
  "family": "devstral",
207
139
  "attachment": false,
208
140
  "reasoning": false,
@@ -230,40 +162,38 @@
230
162
  "output": 2
231
163
  }
232
164
  },
233
- "mistral-large-latest": {
234
- "id": "mistral-large-latest",
235
- "name": "Mistral Large (latest)",
236
- "description": "Flagship Mistral model for advanced reasoning, coding, and multilingual work",
237
- "family": "mistral-large",
238
- "attachment": true,
165
+ "mistral-embed": {
166
+ "id": "mistral-embed",
167
+ "name": "Mistral Embed",
168
+ "description": "Embedding model for semantic search, retrieval, clustering, and ranking pipelines",
169
+ "family": "mistral-embed",
170
+ "attachment": false,
239
171
  "reasoning": false,
240
- "tool_call": true,
241
- "temperature": true,
242
- "knowledge": "2024-11",
243
- "release_date": "2024-11-01",
244
- "last_updated": "2025-12-02",
172
+ "tool_call": false,
173
+ "temperature": false,
174
+ "release_date": "2023-12-11",
175
+ "last_updated": "2023-12-11",
245
176
  "modalities": {
246
177
  "input": [
247
- "text",
248
- "image"
178
+ "text"
249
179
  ],
250
180
  "output": [
251
181
  "text"
252
182
  ]
253
183
  },
254
- "open_weights": true,
184
+ "open_weights": false,
255
185
  "limit": {
256
- "context": 262144,
257
- "output": 262144
186
+ "context": 8000,
187
+ "output": 3072
258
188
  },
259
189
  "cost": {
260
- "input": 0.5,
261
- "output": 1.5
190
+ "input": 0.1,
191
+ "output": 0
262
192
  }
263
193
  },
264
- "devstral-small-2507": {
265
- "id": "devstral-small-2507",
266
- "name": "Devstral Small",
194
+ "devstral-small-2505": {
195
+ "id": "devstral-small-2505",
196
+ "name": "Devstral Small 2505",
267
197
  "description": "Legacy model retained for compatibility with older integrations",
268
198
  "family": "devstral",
269
199
  "attachment": false,
@@ -271,8 +201,8 @@
271
201
  "tool_call": true,
272
202
  "temperature": true,
273
203
  "knowledge": "2025-05",
274
- "release_date": "2025-07-10",
275
- "last_updated": "2025-07-10",
204
+ "release_date": "2025-05-07",
205
+ "last_updated": "2025-05-07",
276
206
  "modalities": {
277
207
  "input": [
278
208
  "text"
@@ -292,37 +222,6 @@
292
222
  "output": 0.3
293
223
  }
294
224
  },
295
- "devstral-2512": {
296
- "id": "devstral-2512",
297
- "name": "Devstral 2",
298
- "description": "Mistral's coding-agent model for repository work, terminal tasks, and software fixes",
299
- "family": "devstral",
300
- "attachment": false,
301
- "reasoning": false,
302
- "tool_call": true,
303
- "temperature": true,
304
- "knowledge": "2025-12",
305
- "release_date": "2025-12-09",
306
- "last_updated": "2025-12-09",
307
- "modalities": {
308
- "input": [
309
- "text"
310
- ],
311
- "output": [
312
- "text"
313
- ]
314
- },
315
- "open_weights": true,
316
- "limit": {
317
- "context": 262144,
318
- "output": 262144
319
- },
320
- "status": "deprecated",
321
- "cost": {
322
- "input": 0.4,
323
- "output": 2
324
- }
325
- },
326
225
  "labs-devstral-small-2512": {
327
226
  "id": "labs-devstral-small-2512",
328
227
  "name": "Devstral Small 2",
@@ -355,18 +254,19 @@
355
254
  "output": 0
356
255
  }
357
256
  },
358
- "open-mistral-nemo": {
359
- "id": "open-mistral-nemo",
360
- "name": "Open Mistral Nemo",
361
- "description": "Legacy model retained for compatibility with older integrations",
362
- "family": "mistral-nemo",
257
+ "magistral-medium-latest": {
258
+ "id": "magistral-medium-latest",
259
+ "name": "Magistral Medium (latest)",
260
+ "description": "Mistral reasoning model for transparent analysis, math, and complex decisions",
261
+ "family": "magistral-medium",
363
262
  "attachment": false,
364
- "reasoning": false,
263
+ "reasoning": true,
264
+ "reasoning_options": [],
365
265
  "tool_call": true,
366
266
  "temperature": true,
367
- "knowledge": "2024-07",
368
- "release_date": "2024-07-01",
369
- "last_updated": "2024-07-01",
267
+ "knowledge": "2025-06",
268
+ "release_date": "2025-03-17",
269
+ "last_updated": "2025-03-20",
370
270
  "modalities": {
371
271
  "input": [
372
272
  "text"
@@ -378,39 +278,28 @@
378
278
  "open_weights": true,
379
279
  "limit": {
380
280
  "context": 128000,
381
- "output": 128000
281
+ "output": 16384
382
282
  },
383
- "status": "deprecated",
384
283
  "cost": {
385
- "input": 0.15,
386
- "output": 0.15
387
- }
284
+ "input": 2,
285
+ "output": 5
286
+ }
388
287
  },
389
- "mistral-small-2603": {
390
- "id": "mistral-small-2603",
391
- "name": "Mistral Small 4",
392
- "description": "Fast Mistral production model for chat, extraction, and cost-sensitive agents",
393
- "family": "mistral-small",
394
- "attachment": true,
395
- "reasoning": true,
396
- "reasoning_options": [
397
- {
398
- "type": "effort",
399
- "values": [
400
- "none",
401
- "high"
402
- ]
403
- }
404
- ],
288
+ "open-mixtral-8x22b": {
289
+ "id": "open-mixtral-8x22b",
290
+ "name": "Mixtral 8x22B",
291
+ "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
292
+ "family": "mixtral",
293
+ "attachment": false,
294
+ "reasoning": false,
405
295
  "tool_call": true,
406
296
  "temperature": true,
407
- "knowledge": "2025-06",
408
- "release_date": "2026-03-16",
409
- "last_updated": "2026-03-16",
297
+ "knowledge": "2024-04",
298
+ "release_date": "2024-04-17",
299
+ "last_updated": "2024-04-17",
410
300
  "modalities": {
411
301
  "input": [
412
- "text",
413
- "image"
302
+ "text"
414
303
  ],
415
304
  "output": [
416
305
  "text"
@@ -418,27 +307,26 @@
418
307
  },
419
308
  "open_weights": true,
420
309
  "limit": {
421
- "context": 256000,
422
- "output": 256000
310
+ "context": 64000,
311
+ "output": 64000
423
312
  },
424
313
  "cost": {
425
- "input": 0.15,
426
- "output": 0.6
314
+ "input": 2,
315
+ "output": 6
427
316
  }
428
317
  },
429
- "magistral-small": {
430
- "id": "magistral-small",
431
- "name": "Magistral Small",
432
- "description": "Mistral reasoning model for transparent analysis, math, and complex decisions",
433
- "family": "magistral-small",
318
+ "open-mixtral-8x7b": {
319
+ "id": "open-mixtral-8x7b",
320
+ "name": "Mixtral 8x7B",
321
+ "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
322
+ "family": "mixtral",
434
323
  "attachment": false,
435
- "reasoning": true,
436
- "reasoning_options": [],
324
+ "reasoning": false,
437
325
  "tool_call": true,
438
326
  "temperature": true,
439
- "knowledge": "2025-06",
440
- "release_date": "2025-03-17",
441
- "last_updated": "2025-03-17",
327
+ "knowledge": "2024-01",
328
+ "release_date": "2023-12-11",
329
+ "last_updated": "2023-12-11",
442
330
  "modalities": {
443
331
  "input": [
444
332
  "text"
@@ -449,51 +337,65 @@
449
337
  },
450
338
  "open_weights": true,
451
339
  "limit": {
452
- "context": 128000,
453
- "output": 128000
340
+ "context": 32000,
341
+ "output": 32000
454
342
  },
455
343
  "cost": {
456
- "input": 0.5,
457
- "output": 1.5
344
+ "input": 0.7,
345
+ "output": 0.7
458
346
  }
459
347
  },
460
- "voxtral-mini-latest": {
461
- "id": "voxtral-mini-latest",
462
- "name": "Voxtral Mini (latest)",
463
- "description": "Speech transcription model for accurate audio-to-text and captioning workflows",
464
- "family": "voxtral",
348
+ "open-mistral-7b": {
349
+ "id": "open-mistral-7b",
350
+ "name": "Mistral 7B",
351
+ "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
352
+ "family": "mistral",
465
353
  "attachment": false,
466
354
  "reasoning": false,
467
- "tool_call": false,
468
- "temperature": false,
469
- "release_date": "2026-02-01",
470
- "last_updated": "2026-02-01",
355
+ "tool_call": true,
356
+ "temperature": true,
357
+ "knowledge": "2023-12",
358
+ "release_date": "2023-09-27",
359
+ "last_updated": "2023-09-27",
471
360
  "modalities": {
472
361
  "input": [
473
- "audio"
362
+ "text"
474
363
  ],
475
364
  "output": [
476
365
  "text"
477
366
  ]
478
367
  },
479
- "open_weights": false,
368
+ "open_weights": true,
480
369
  "limit": {
481
- "context": 0,
482
- "output": 0
370
+ "context": 8000,
371
+ "output": 8000
372
+ },
373
+ "cost": {
374
+ "input": 0.25,
375
+ "output": 0.25
483
376
  }
484
377
  },
485
- "mistral-small-2506": {
486
- "id": "mistral-small-2506",
487
- "name": "Mistral Small 3.2",
488
- "description": "Efficient Mistral model for fast chat, extraction, and production assistants",
489
- "family": "mistral-small",
490
- "attachment": false,
491
- "reasoning": false,
378
+ "mistral-medium-latest": {
379
+ "id": "mistral-medium-latest",
380
+ "name": "Mistral Medium (latest)",
381
+ "description": "Balanced Mistral model for enterprise assistants, multilingual work, and tools",
382
+ "family": "mistral-medium",
383
+ "attachment": true,
384
+ "reasoning": true,
385
+ "reasoning_options": [
386
+ {
387
+ "type": "effort",
388
+ "values": [
389
+ "none",
390
+ "high"
391
+ ]
392
+ }
393
+ ],
492
394
  "tool_call": true,
395
+ "structured_output": true,
493
396
  "temperature": true,
494
- "knowledge": "2025-03",
495
- "release_date": "2025-06-20",
496
- "last_updated": "2025-06-20",
397
+ "release_date": "2026-04-29",
398
+ "last_updated": "2026-04-29",
497
399
  "modalities": {
498
400
  "input": [
499
401
  "text",
@@ -504,27 +406,67 @@
504
406
  ]
505
407
  },
506
408
  "open_weights": true,
409
+ "limit": {
410
+ "context": 262144,
411
+ "output": 262144
412
+ },
413
+ "cost": {
414
+ "input": 1.5,
415
+ "output": 7.5
416
+ }
417
+ },
418
+ "devstral-medium-2507": {
419
+ "id": "devstral-medium-2507",
420
+ "name": "Devstral Medium",
421
+ "description": "Legacy model retained for compatibility with older integrations",
422
+ "family": "devstral",
423
+ "attachment": false,
424
+ "reasoning": false,
425
+ "tool_call": true,
426
+ "temperature": true,
427
+ "knowledge": "2025-05",
428
+ "release_date": "2025-07-10",
429
+ "last_updated": "2025-07-10",
430
+ "modalities": {
431
+ "input": [
432
+ "text"
433
+ ],
434
+ "output": [
435
+ "text"
436
+ ]
437
+ },
438
+ "open_weights": true,
507
439
  "limit": {
508
440
  "context": 128000,
509
- "output": 16384
441
+ "output": 128000
510
442
  },
443
+ "status": "deprecated",
511
444
  "cost": {
512
- "input": 0.1,
513
- "output": 0.3
445
+ "input": 0.4,
446
+ "output": 2
514
447
  }
515
448
  },
516
- "pixtral-large-latest": {
517
- "id": "pixtral-large-latest",
518
- "name": "Pixtral Large (latest)",
519
- "description": "Mistral's larger vision model for document-heavy image understanding and chat",
520
- "family": "pixtral",
449
+ "mistral-medium-2604": {
450
+ "id": "mistral-medium-2604",
451
+ "name": "Mistral Medium 3.5",
452
+ "description": "Balanced Mistral model for enterprise assistants, multilingual work, and tools",
453
+ "family": "mistral-medium",
521
454
  "attachment": true,
522
- "reasoning": false,
455
+ "reasoning": true,
456
+ "reasoning_options": [
457
+ {
458
+ "type": "effort",
459
+ "values": [
460
+ "none",
461
+ "high"
462
+ ]
463
+ }
464
+ ],
523
465
  "tool_call": true,
466
+ "structured_output": true,
524
467
  "temperature": true,
525
- "knowledge": "2024-11",
526
- "release_date": "2024-11-01",
527
- "last_updated": "2024-11-04",
468
+ "release_date": "2026-04-29",
469
+ "last_updated": "2026-04-29",
528
470
  "modalities": {
529
471
  "input": [
530
472
  "text",
@@ -536,12 +478,12 @@
536
478
  },
537
479
  "open_weights": true,
538
480
  "limit": {
539
- "context": 128000,
540
- "output": 128000
481
+ "context": 262144,
482
+ "output": 262144
541
483
  },
542
484
  "cost": {
543
- "input": 2,
544
- "output": 6
485
+ "input": 1.5,
486
+ "output": 7.5
545
487
  }
546
488
  },
547
489
  "mistral-large-2512": {
@@ -575,9 +517,70 @@
575
517
  "output": 1.5
576
518
  }
577
519
  },
578
- "ministral-3b-latest": {
579
- "id": "ministral-3b-latest",
580
- "name": "Ministral 3B (latest)",
520
+ "devstral-medium-latest": {
521
+ "id": "devstral-medium-latest",
522
+ "name": "Devstral 2 (latest)",
523
+ "description": "Legacy model retained for compatibility with older integrations",
524
+ "family": "devstral",
525
+ "attachment": false,
526
+ "reasoning": false,
527
+ "tool_call": true,
528
+ "temperature": true,
529
+ "knowledge": "2025-12",
530
+ "release_date": "2025-12-02",
531
+ "last_updated": "2025-12-02",
532
+ "modalities": {
533
+ "input": [
534
+ "text"
535
+ ],
536
+ "output": [
537
+ "text"
538
+ ]
539
+ },
540
+ "open_weights": true,
541
+ "limit": {
542
+ "context": 262144,
543
+ "output": 262144
544
+ },
545
+ "status": "deprecated",
546
+ "cost": {
547
+ "input": 0.4,
548
+ "output": 2
549
+ }
550
+ },
551
+ "voxtral-small-latest": {
552
+ "id": "voxtral-small-latest",
553
+ "name": "Voxtral Small (latest)",
554
+ "description": "Instruct model with native audio input for speech understanding and tool use",
555
+ "family": "voxtral",
556
+ "attachment": true,
557
+ "reasoning": false,
558
+ "tool_call": true,
559
+ "temperature": true,
560
+ "release_date": "2025-07-15",
561
+ "last_updated": "2025-07-15",
562
+ "modalities": {
563
+ "input": [
564
+ "text",
565
+ "audio"
566
+ ],
567
+ "output": [
568
+ "text"
569
+ ]
570
+ },
571
+ "open_weights": true,
572
+ "limit": {
573
+ "context": 32000,
574
+ "output": 32000
575
+ },
576
+ "cost": {
577
+ "input": 0.1,
578
+ "output": 0.3
579
+ }
580
+ },
581
+ "ministral-8b-latest": {
582
+ "id": "ministral-8b-latest",
583
+ "name": "Ministral 8B (latest)",
581
584
  "description": "Compact Mistral model for edge, latency-sensitive, and cost-efficient workloads",
582
585
  "family": "ministral",
583
586
  "attachment": false,
@@ -601,15 +604,70 @@
601
604
  "output": 128000
602
605
  },
603
606
  "cost": {
604
- "input": 0.04,
605
- "output": 0.04
607
+ "input": 0.1,
608
+ "output": 0.1
606
609
  }
607
610
  },
608
- "mistral-medium-2604": {
609
- "id": "mistral-medium-2604",
610
- "name": "Mistral Medium 3.5",
611
- "description": "Balanced Mistral model for enterprise assistants, multilingual work, and tools",
612
- "family": "mistral-medium",
611
+ "mistral-nemo": {
612
+ "id": "mistral-nemo",
613
+ "name": "Mistral Nemo",
614
+ "description": "Efficient Mistral-NVIDIA open model for multilingual chat and local deployment",
615
+ "family": "mistral-nemo",
616
+ "attachment": false,
617
+ "reasoning": false,
618
+ "tool_call": true,
619
+ "temperature": true,
620
+ "knowledge": "2024-07",
621
+ "release_date": "2024-07-01",
622
+ "last_updated": "2024-07-01",
623
+ "modalities": {
624
+ "input": [
625
+ "text"
626
+ ],
627
+ "output": [
628
+ "text"
629
+ ]
630
+ },
631
+ "open_weights": true,
632
+ "limit": {
633
+ "context": 128000,
634
+ "output": 128000
635
+ },
636
+ "cost": {
637
+ "input": 0.15,
638
+ "output": 0.15
639
+ }
640
+ },
641
+ "voxtral-mini-tts-latest": {
642
+ "id": "voxtral-mini-tts-latest",
643
+ "name": "Voxtral Mini TTS (latest)",
644
+ "description": "Multilingual text-to-speech model with zero-shot voice cloning",
645
+ "family": "voxtral",
646
+ "attachment": false,
647
+ "reasoning": false,
648
+ "tool_call": false,
649
+ "temperature": false,
650
+ "release_date": "2026-03-01",
651
+ "last_updated": "2026-03-01",
652
+ "modalities": {
653
+ "input": [
654
+ "text"
655
+ ],
656
+ "output": [
657
+ "audio"
658
+ ]
659
+ },
660
+ "open_weights": false,
661
+ "limit": {
662
+ "context": 0,
663
+ "output": 0
664
+ }
665
+ },
666
+ "mistral-small-latest": {
667
+ "id": "mistral-small-latest",
668
+ "name": "Mistral Small (latest)",
669
+ "description": "Efficient Mistral model for fast chat, extraction, and production assistants",
670
+ "family": "mistral-small",
613
671
  "attachment": true,
614
672
  "reasoning": true,
615
673
  "reasoning_options": [
@@ -622,10 +680,10 @@
622
680
  }
623
681
  ],
624
682
  "tool_call": true,
625
- "structured_output": true,
626
683
  "temperature": true,
627
- "release_date": "2026-04-29",
628
- "last_updated": "2026-04-29",
684
+ "knowledge": "2025-06",
685
+ "release_date": "2026-03-16",
686
+ "last_updated": "2026-03-16",
629
687
  "modalities": {
630
688
  "input": [
631
689
  "text",
@@ -637,26 +695,26 @@
637
695
  },
638
696
  "open_weights": true,
639
697
  "limit": {
640
- "context": 262144,
641
- "output": 262144
698
+ "context": 256000,
699
+ "output": 256000
642
700
  },
643
701
  "cost": {
644
- "input": 1.5,
645
- "output": 7.5
702
+ "input": 0.15,
703
+ "output": 0.6
646
704
  }
647
705
  },
648
- "devstral-small-2505": {
649
- "id": "devstral-small-2505",
650
- "name": "Devstral Small 2505",
706
+ "open-mistral-nemo": {
707
+ "id": "open-mistral-nemo",
708
+ "name": "Open Mistral Nemo",
651
709
  "description": "Legacy model retained for compatibility with older integrations",
652
- "family": "devstral",
710
+ "family": "mistral-nemo",
653
711
  "attachment": false,
654
712
  "reasoning": false,
655
713
  "tool_call": true,
656
714
  "temperature": true,
657
- "knowledge": "2025-05",
658
- "release_date": "2025-05-07",
659
- "last_updated": "2025-05-07",
715
+ "knowledge": "2024-07",
716
+ "release_date": "2024-07-01",
717
+ "last_updated": "2024-07-01",
660
718
  "modalities": {
661
719
  "input": [
662
720
  "text"
@@ -672,25 +730,26 @@
672
730
  },
673
731
  "status": "deprecated",
674
732
  "cost": {
675
- "input": 0.1,
676
- "output": 0.3
733
+ "input": 0.15,
734
+ "output": 0.15
677
735
  }
678
736
  },
679
- "ministral-8b-latest": {
680
- "id": "ministral-8b-latest",
681
- "name": "Ministral 8B (latest)",
682
- "description": "Compact Mistral model for edge, latency-sensitive, and cost-efficient workloads",
683
- "family": "ministral",
684
- "attachment": false,
737
+ "mistral-large-latest": {
738
+ "id": "mistral-large-latest",
739
+ "name": "Mistral Large (latest)",
740
+ "description": "Flagship Mistral model for advanced reasoning, coding, and multilingual work",
741
+ "family": "mistral-large",
742
+ "attachment": true,
685
743
  "reasoning": false,
686
744
  "tool_call": true,
687
745
  "temperature": true,
688
- "knowledge": "2024-10",
689
- "release_date": "2024-10-01",
690
- "last_updated": "2024-10-04",
746
+ "knowledge": "2024-11",
747
+ "release_date": "2024-11-01",
748
+ "last_updated": "2025-12-02",
691
749
  "modalities": {
692
750
  "input": [
693
- "text"
751
+ "text",
752
+ "image"
694
753
  ],
695
754
  "output": [
696
755
  "text"
@@ -698,48 +757,72 @@
698
757
  },
699
758
  "open_weights": true,
700
759
  "limit": {
701
- "context": 128000,
702
- "output": 128000
760
+ "context": 262144,
761
+ "output": 262144
703
762
  },
704
763
  "cost": {
705
- "input": 0.1,
706
- "output": 0.1
764
+ "input": 0.5,
765
+ "output": 1.5
707
766
  }
708
767
  },
709
- "mistral-medium-2505": {
710
- "id": "mistral-medium-2505",
711
- "name": "Mistral Medium 3",
712
- "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
713
- "family": "mistral-medium",
714
- "attachment": true,
768
+ "voxtral-mini-latest": {
769
+ "id": "voxtral-mini-latest",
770
+ "name": "Voxtral Mini (latest)",
771
+ "description": "Speech transcription model for accurate audio-to-text and captioning workflows",
772
+ "family": "voxtral",
773
+ "attachment": false,
774
+ "reasoning": false,
775
+ "tool_call": false,
776
+ "temperature": false,
777
+ "release_date": "2026-02-01",
778
+ "last_updated": "2026-02-01",
779
+ "modalities": {
780
+ "input": [
781
+ "audio"
782
+ ],
783
+ "output": [
784
+ "text"
785
+ ]
786
+ },
787
+ "open_weights": false,
788
+ "limit": {
789
+ "context": 0,
790
+ "output": 0
791
+ }
792
+ },
793
+ "mistral-large-2411": {
794
+ "id": "mistral-large-2411",
795
+ "name": "Mistral Large 2.1",
796
+ "description": "Flagship Mistral model for advanced reasoning, coding, and multilingual work",
797
+ "family": "mistral-large",
798
+ "attachment": false,
715
799
  "reasoning": false,
716
800
  "tool_call": true,
717
801
  "temperature": true,
718
- "knowledge": "2025-05",
719
- "release_date": "2025-05-07",
720
- "last_updated": "2025-05-07",
802
+ "knowledge": "2024-11",
803
+ "release_date": "2024-11-18",
804
+ "last_updated": "2024-11-18",
721
805
  "modalities": {
722
806
  "input": [
723
- "text",
724
- "image"
807
+ "text"
725
808
  ],
726
809
  "output": [
727
810
  "text"
728
811
  ]
729
812
  },
730
- "open_weights": false,
813
+ "open_weights": true,
731
814
  "limit": {
732
815
  "context": 131072,
733
- "output": 131072
816
+ "output": 16384
734
817
  },
735
818
  "cost": {
736
- "input": 0.4,
737
- "output": 2
819
+ "input": 2,
820
+ "output": 6
738
821
  }
739
822
  },
740
- "devstral-medium-latest": {
741
- "id": "devstral-medium-latest",
742
- "name": "Devstral 2 (latest)",
823
+ "devstral-latest": {
824
+ "id": "devstral-latest",
825
+ "name": "Devstral 2",
743
826
  "description": "Legacy model retained for compatibility with older integrations",
744
827
  "family": "devstral",
745
828
  "attachment": false,
@@ -747,8 +830,8 @@
747
830
  "tool_call": true,
748
831
  "temperature": true,
749
832
  "knowledge": "2025-12",
750
- "release_date": "2025-12-02",
751
- "last_updated": "2025-12-02",
833
+ "release_date": "2025-12-09",
834
+ "last_updated": "2025-12-09",
752
835
  "modalities": {
753
836
  "input": [
754
837
  "text"
@@ -768,18 +851,27 @@
768
851
  "output": 2
769
852
  }
770
853
  },
771
- "open-mistral-7b": {
772
- "id": "open-mistral-7b",
773
- "name": "Mistral 7B",
774
- "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
775
- "family": "mistral",
854
+ "zai-glm-5-2": {
855
+ "id": "zai-glm-5-2",
856
+ "name": "GLM-5.2",
857
+ "description": "Open flagship GLM for long-horizon coding agents and million-token context work",
858
+ "family": "glm",
776
859
  "attachment": false,
777
- "reasoning": false,
860
+ "reasoning": true,
861
+ "reasoning_options": [
862
+ {
863
+ "type": "effort",
864
+ "values": [
865
+ "high",
866
+ "max"
867
+ ]
868
+ }
869
+ ],
778
870
  "tool_call": true,
871
+ "structured_output": true,
779
872
  "temperature": true,
780
- "knowledge": "2023-12",
781
- "release_date": "2023-09-27",
782
- "last_updated": "2023-09-27",
873
+ "release_date": "2026-06-13",
874
+ "last_updated": "2026-06-13",
783
875
  "modalities": {
784
876
  "input": [
785
877
  "text"
@@ -790,51 +882,37 @@
790
882
  },
791
883
  "open_weights": true,
792
884
  "limit": {
793
- "context": 8000,
794
- "output": 8000
885
+ "context": 1000000,
886
+ "output": 131072
795
887
  },
888
+ "status": "beta",
796
889
  "cost": {
797
- "input": 0.25,
798
- "output": 0.25
799
- }
800
- },
801
- "voxtral-mini-tts-latest": {
802
- "id": "voxtral-mini-tts-latest",
803
- "name": "Voxtral Mini TTS (latest)",
804
- "description": "Multilingual text-to-speech model with zero-shot voice cloning",
805
- "family": "voxtral",
806
- "attachment": false,
807
- "reasoning": false,
808
- "tool_call": false,
809
- "temperature": false,
810
- "release_date": "2026-03-01",
811
- "last_updated": "2026-03-01",
812
- "modalities": {
813
- "input": [
814
- "text"
815
- ],
816
- "output": [
817
- "audio"
818
- ]
819
- },
820
- "open_weights": false,
821
- "limit": {
822
- "context": 0,
823
- "output": 0
890
+ "input": 1.4,
891
+ "output": 4.4,
892
+ "cache_read": 0.14
824
893
  }
825
894
  },
826
- "pixtral-12b": {
827
- "id": "pixtral-12b",
828
- "name": "Pixtral 12B",
829
- "description": "Mistral vision-language model for image understanding and multimodal chat",
830
- "family": "pixtral",
895
+ "mistral-small-2603": {
896
+ "id": "mistral-small-2603",
897
+ "name": "Mistral Small 4",
898
+ "description": "Fast Mistral production model for chat, extraction, and cost-sensitive agents",
899
+ "family": "mistral-small",
831
900
  "attachment": true,
832
- "reasoning": false,
901
+ "reasoning": true,
902
+ "reasoning_options": [
903
+ {
904
+ "type": "effort",
905
+ "values": [
906
+ "none",
907
+ "high"
908
+ ]
909
+ }
910
+ ],
833
911
  "tool_call": true,
834
912
  "temperature": true,
835
- "knowledge": "2024-09",
836
- "release_date": "2024-09-01",
837
- "last_updated": "2024-09-01",
913
+ "knowledge": "2025-06",
914
+ "release_date": "2026-03-16",
915
+ "last_updated": "2026-03-16",
838
916
  "modalities": {
839
917
  "input": [
840
918
  "text",
@@ -846,42 +924,43 @@
846
924
  },
847
925
  "open_weights": true,
848
926
  "limit": {
849
- "context": 128000,
850
- "output": 128000
927
+ "context": 256000,
928
+ "output": 256000
851
929
  },
852
930
  "cost": {
853
931
  "input": 0.15,
854
- "output": 0.15
932
+ "output": 0.6
855
933
  }
856
934
  },
857
- "voxtral-small-latest": {
858
- "id": "voxtral-small-latest",
859
- "name": "Voxtral Small (latest)",
860
- "description": "Instruct model with native audio input for speech understanding and tool use",
861
- "family": "voxtral",
935
+ "mistral-medium-2505": {
936
+ "id": "mistral-medium-2505",
937
+ "name": "Mistral Medium 3",
938
+ "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
939
+ "family": "mistral-medium",
862
940
  "attachment": true,
863
941
  "reasoning": false,
864
942
  "tool_call": true,
865
943
  "temperature": true,
866
- "release_date": "2025-07-15",
867
- "last_updated": "2025-07-15",
944
+ "knowledge": "2025-05",
945
+ "release_date": "2025-05-07",
946
+ "last_updated": "2025-05-07",
868
947
  "modalities": {
869
948
  "input": [
870
949
  "text",
871
- "audio"
950
+ "image"
872
951
  ],
873
952
  "output": [
874
953
  "text"
875
954
  ]
876
955
  },
877
- "open_weights": true,
956
+ "open_weights": false,
878
957
  "limit": {
879
- "context": 32000,
880
- "output": 32000
958
+ "context": 131072,
959
+ "output": 131072
881
960
  },
882
961
  "cost": {
883
- "input": 0.1,
884
- "output": 0.3
962
+ "input": 0.4,
963
+ "output": 2
885
964
  }
886
965
  },
887
966
  "codestral-latest": {
@@ -914,49 +993,18 @@
914
993
  "output": 0.9
915
994
  }
916
995
  },
917
- "open-mixtral-8x22b": {
918
- "id": "open-mixtral-8x22b",
919
- "name": "Mixtral 8x22B",
920
- "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
921
- "family": "mixtral",
996
+ "ministral-3b-latest": {
997
+ "id": "ministral-3b-latest",
998
+ "name": "Ministral 3B (latest)",
999
+ "description": "Compact Mistral model for edge, latency-sensitive, and cost-efficient workloads",
1000
+ "family": "ministral",
922
1001
  "attachment": false,
923
1002
  "reasoning": false,
924
1003
  "tool_call": true,
925
1004
  "temperature": true,
926
- "knowledge": "2024-04",
927
- "release_date": "2024-04-17",
928
- "last_updated": "2024-04-17",
929
- "modalities": {
930
- "input": [
931
- "text"
932
- ],
933
- "output": [
934
- "text"
935
- ]
936
- },
937
- "open_weights": true,
938
- "limit": {
939
- "context": 64000,
940
- "output": 64000
941
- },
942
- "cost": {
943
- "input": 2,
944
- "output": 6
945
- }
946
- },
947
- "magistral-medium-latest": {
948
- "id": "magistral-medium-latest",
949
- "name": "Magistral Medium (latest)",
950
- "description": "Mistral reasoning model for transparent analysis, math, and complex decisions",
951
- "family": "magistral-medium",
952
- "attachment": false,
953
- "reasoning": true,
954
- "reasoning_options": [],
955
- "tool_call": true,
956
- "temperature": true,
957
- "knowledge": "2025-06",
958
- "release_date": "2025-03-17",
959
- "last_updated": "2025-03-20",
1005
+ "knowledge": "2024-10",
1006
+ "release_date": "2024-10-01",
1007
+ "last_updated": "2024-10-04",
960
1008
  "modalities": {
961
1009
  "input": [
962
1010
  "text"
@@ -968,27 +1016,29 @@
968
1016
  "open_weights": true,
969
1017
  "limit": {
970
1018
  "context": 128000,
971
- "output": 16384
1019
+ "output": 128000
972
1020
  },
973
1021
  "cost": {
974
- "input": 2,
975
- "output": 5
1022
+ "input": 0.04,
1023
+ "output": 0.04
976
1024
  }
977
1025
  },
978
- "mistral-embed": {
979
- "id": "mistral-embed",
980
- "name": "Mistral Embed",
981
- "description": "Embedding model for semantic search, retrieval, clustering, and ranking pipelines",
982
- "family": "mistral-embed",
983
- "attachment": false,
1026
+ "mistral-medium-2508": {
1027
+ "id": "mistral-medium-2508",
1028
+ "name": "Mistral Medium 3.1",
1029
+ "description": "Mistral model for multilingual chat, reasoning, and tool-assisted workflows",
1030
+ "family": "mistral-medium",
1031
+ "attachment": true,
984
1032
  "reasoning": false,
985
- "tool_call": false,
986
- "temperature": false,
987
- "release_date": "2023-12-11",
988
- "last_updated": "2023-12-11",
1033
+ "tool_call": true,
1034
+ "temperature": true,
1035
+ "knowledge": "2025-05",
1036
+ "release_date": "2025-08-12",
1037
+ "last_updated": "2025-08-12",
989
1038
  "modalities": {
990
1039
  "input": [
991
- "text"
1040
+ "text",
1041
+ "image"
992
1042
  ],
993
1043
  "output": [
994
1044
  "text"
@@ -996,35 +1046,26 @@
996
1046
  },
997
1047
  "open_weights": false,
998
1048
  "limit": {
999
- "context": 8000,
1000
- "output": 3072
1049
+ "context": 262144,
1050
+ "output": 262144
1001
1051
  },
1002
1052
  "cost": {
1003
- "input": 0.1,
1004
- "output": 0
1053
+ "input": 0.4,
1054
+ "output": 2
1005
1055
  }
1006
1056
  },
1007
- "mistral-medium-latest": {
1008
- "id": "mistral-medium-latest",
1009
- "name": "Mistral Medium (latest)",
1010
- "description": "Balanced Mistral model for enterprise assistants, multilingual work, and tools",
1011
- "family": "mistral-medium",
1057
+ "pixtral-large-latest": {
1058
+ "id": "pixtral-large-latest",
1059
+ "name": "Pixtral Large (latest)",
1060
+ "description": "Mistral's larger vision model for document-heavy image understanding and chat",
1061
+ "family": "pixtral",
1012
1062
  "attachment": true,
1013
- "reasoning": true,
1014
- "reasoning_options": [
1015
- {
1016
- "type": "effort",
1017
- "values": [
1018
- "none",
1019
- "high"
1020
- ]
1021
- }
1022
- ],
1063
+ "reasoning": false,
1023
1064
  "tool_call": true,
1024
- "structured_output": true,
1025
1065
  "temperature": true,
1026
- "release_date": "2026-04-29",
1027
- "last_updated": "2026-04-29",
1066
+ "knowledge": "2024-11",
1067
+ "release_date": "2024-11-01",
1068
+ "last_updated": "2024-11-04",
1028
1069
  "modalities": {
1029
1070
  "input": [
1030
1071
  "text",
@@ -1036,12 +1077,12 @@
1036
1077
  },
1037
1078
  "open_weights": true,
1038
1079
  "limit": {
1039
- "context": 262144,
1040
- "output": 262144
1080
+ "context": 128000,
1081
+ "output": 128000
1041
1082
  },
1042
1083
  "cost": {
1043
- "input": 1.5,
1044
- "output": 7.5
1084
+ "input": 2,
1085
+ "output": 6
1045
1086
  }
1046
1087
  }
1047
1088
  }