llm.rb 11.3.1 → 12.1.0

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 (76) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGELOG.md +354 -2
  3. data/LICENSE +93 -17
  4. data/README.md +226 -616
  5. data/data/anthropic.json +322 -426
  6. data/data/bedrock.json +2634 -1144
  7. data/data/deepinfra.json +1513 -0
  8. data/data/deepseek.json +57 -28
  9. data/data/google.json +411 -771
  10. data/data/openai.json +1104 -771
  11. data/data/xai.json +141 -292
  12. data/data/zai.json +263 -141
  13. data/lib/llm/active_record/acts_as_agent.rb +3 -41
  14. data/lib/llm/active_record/acts_as_llm.rb +18 -0
  15. data/lib/llm/active_record.rb +3 -3
  16. data/lib/llm/agent.rb +25 -1
  17. data/lib/llm/context.rb +17 -5
  18. data/lib/llm/contract/completion.rb +2 -2
  19. data/lib/llm/json_adapter.rb +29 -3
  20. data/lib/llm/provider.rb +3 -3
  21. data/lib/llm/providers/deepinfra/audio.rb +66 -0
  22. data/lib/llm/providers/deepinfra/images.rb +90 -0
  23. data/lib/llm/providers/deepinfra/response_adapter.rb +36 -0
  24. data/lib/llm/providers/deepinfra.rb +100 -0
  25. data/lib/llm/providers/deepseek/images.rb +109 -0
  26. data/lib/llm/providers/deepseek/request_adapter.rb +32 -0
  27. data/lib/llm/providers/deepseek/response_adapter/image.rb +9 -0
  28. data/lib/llm/providers/deepseek/response_adapter.rb +29 -0
  29. data/lib/llm/providers/deepseek.rb +4 -2
  30. data/lib/llm/providers/google/request_adapter.rb +22 -5
  31. data/lib/llm/providers/google.rb +4 -4
  32. data/lib/llm/providers/llamacpp.rb +5 -5
  33. data/lib/llm/providers/openai/audio.rb +6 -2
  34. data/lib/llm/providers/openai/images.rb +9 -50
  35. data/lib/llm/providers/openai/request_adapter/respond.rb +38 -4
  36. data/lib/llm/providers/openai/response_adapter/audio.rb +5 -1
  37. data/lib/llm/providers/openai/response_adapter/completion.rb +1 -1
  38. data/lib/llm/providers/openai/response_adapter/image.rb +0 -4
  39. data/lib/llm/providers/openai/responses.rb +1 -0
  40. data/lib/llm/providers/openai/stream_parser.rb +5 -6
  41. data/lib/llm/providers/openai.rb +2 -2
  42. data/lib/llm/providers/xai/images.rb +49 -26
  43. data/lib/llm/providers/xai.rb +2 -2
  44. data/lib/llm/repl/input.rb +64 -0
  45. data/lib/llm/repl/status.rb +30 -0
  46. data/lib/llm/repl/stream.rb +46 -0
  47. data/lib/llm/repl/transcript.rb +61 -0
  48. data/lib/llm/repl/window.rb +107 -0
  49. data/lib/llm/repl.rb +78 -0
  50. data/lib/llm/response.rb +10 -0
  51. data/lib/llm/schema/leaf.rb +7 -1
  52. data/lib/llm/schema/renderer.rb +121 -0
  53. data/lib/llm/schema.rb +30 -0
  54. data/lib/llm/sequel/agent.rb +2 -43
  55. data/lib/llm/sequel/plugin.rb +25 -7
  56. data/lib/llm/tools/chdir.rb +23 -0
  57. data/lib/llm/tools/git.rb +41 -0
  58. data/lib/llm/tools/mkdir.rb +32 -0
  59. data/lib/llm/tools/pwd.rb +20 -0
  60. data/lib/llm/tools/read_file.rb +40 -0
  61. data/lib/llm/tools/rg.rb +46 -0
  62. data/lib/llm/tools/shell.rb +48 -0
  63. data/lib/llm/tools/swap_text.rb +25 -0
  64. data/lib/llm/tools/write_file.rb +24 -0
  65. data/lib/llm/tools.rb +5 -0
  66. data/lib/llm/tracer/telemetry.rb +4 -6
  67. data/lib/llm/tracer.rb +9 -21
  68. data/lib/llm/transport/execution.rb +16 -1
  69. data/lib/llm/transport/net_http_adapter.rb +1 -1
  70. data/lib/llm/uridata.rb +16 -0
  71. data/lib/llm/version.rb +1 -1
  72. data/lib/llm.rb +9 -0
  73. data/llm.gemspec +5 -18
  74. data/resources/deepdive.md +829 -263
  75. metadata +31 -18
  76. data/lib/llm/tracer/langsmith.rb +0 -144
data/data/google.json CHANGED
@@ -1,6 +1,7 @@
1
1
  {
2
2
  "id": "google",
3
3
  "env": [
4
+ "GOOGLE_API_KEY",
4
5
  "GOOGLE_GENERATIVE_AI_API_KEY",
5
6
  "GEMINI_API_KEY"
6
7
  ],
@@ -8,107 +9,24 @@
8
9
  "name": "Google",
9
10
  "doc": "https://ai.google.dev/gemini-api/docs/models",
10
11
  "models": {
11
- "gemini-flash-latest": {
12
- "id": "gemini-flash-latest",
13
- "name": "Gemini Flash Latest",
14
- "family": "gemini-flash",
15
- "attachment": true,
16
- "reasoning": true,
17
- "tool_call": true,
18
- "structured_output": true,
19
- "temperature": true,
20
- "knowledge": "2025-01",
21
- "release_date": "2025-09-25",
22
- "last_updated": "2025-09-25",
23
- "modalities": {
24
- "input": [
25
- "text",
26
- "image",
27
- "audio",
28
- "video",
29
- "pdf"
30
- ],
31
- "output": [
32
- "text"
33
- ]
34
- },
35
- "open_weights": false,
36
- "limit": {
37
- "context": 1048576,
38
- "output": 65536
39
- },
40
- "cost": {
41
- "input": 0.3,
42
- "output": 2.5,
43
- "cache_read": 0.075,
44
- "input_audio": 1
45
- }
46
- },
47
- "gemma-4-26b-a4b-it": {
48
- "id": "gemma-4-26b-a4b-it",
49
- "name": "Gemma 4 26B",
50
- "family": "gemma",
51
- "attachment": false,
52
- "reasoning": true,
53
- "tool_call": true,
54
- "structured_output": true,
55
- "temperature": true,
56
- "release_date": "2026-04-02",
57
- "last_updated": "2026-04-02",
58
- "modalities": {
59
- "input": [
60
- "text",
61
- "image"
62
- ],
63
- "output": [
64
- "text"
65
- ]
66
- },
67
- "open_weights": true,
68
- "limit": {
69
- "context": 256000,
70
- "output": 8192
71
- }
72
- },
73
- "gemini-1.5-flash": {
74
- "id": "gemini-1.5-flash",
75
- "name": "Gemini 1.5 Flash",
76
- "family": "gemini-flash",
77
- "attachment": true,
78
- "reasoning": false,
79
- "tool_call": true,
80
- "temperature": true,
81
- "knowledge": "2024-04",
82
- "release_date": "2024-05-14",
83
- "last_updated": "2024-05-14",
84
- "modalities": {
85
- "input": [
86
- "text",
87
- "image",
88
- "audio",
89
- "video"
90
- ],
91
- "output": [
92
- "text"
93
- ]
94
- },
95
- "open_weights": false,
96
- "limit": {
97
- "context": 1000000,
98
- "output": 8192
99
- },
100
- "cost": {
101
- "input": 0.075,
102
- "output": 0.3,
103
- "cache_read": 0.01875
104
- }
105
- },
106
12
  "gemini-3.1-flash-lite": {
107
13
  "id": "gemini-3.1-flash-lite",
108
14
  "name": "Gemini 3.1 Flash Lite",
15
+ "description": "Low-latency Gemini model for high-volume multimodal and agent workloads",
109
16
  "family": "gemini-flash-lite",
110
17
  "attachment": true,
111
18
  "reasoning": true,
19
+ "reasoning_options": [
20
+ {
21
+ "type": "effort",
22
+ "values": [
23
+ "minimal",
24
+ "low",
25
+ "medium",
26
+ "high"
27
+ ]
28
+ }
29
+ ],
112
30
  "tool_call": true,
113
31
  "structured_output": true,
114
32
  "temperature": true,
@@ -139,54 +57,56 @@
139
57
  "input_audio": 0.5
140
58
  }
141
59
  },
142
- "gemini-2.5-flash-preview-09-2025": {
143
- "id": "gemini-2.5-flash-preview-09-2025",
144
- "name": "Gemini 2.5 Flash Preview 09-25",
60
+ "gemini-2.5-flash-preview-tts": {
61
+ "id": "gemini-2.5-flash-preview-tts",
62
+ "name": "Gemini 2.5 Flash Preview TTS",
63
+ "description": "Speech generation model for controllable voice, narration, and audio delivery",
145
64
  "family": "gemini-flash",
146
- "attachment": true,
147
- "reasoning": true,
148
- "tool_call": true,
149
- "structured_output": true,
65
+ "attachment": false,
66
+ "reasoning": false,
67
+ "tool_call": false,
150
68
  "temperature": true,
151
69
  "knowledge": "2025-01",
152
- "release_date": "2025-09-25",
153
- "last_updated": "2025-09-25",
70
+ "release_date": "2025-05-01",
71
+ "last_updated": "2025-05-01",
154
72
  "modalities": {
155
73
  "input": [
156
- "text",
157
- "image",
158
- "audio",
159
- "video",
160
- "pdf"
74
+ "text"
161
75
  ],
162
76
  "output": [
163
- "text"
77
+ "audio"
164
78
  ]
165
79
  },
166
80
  "open_weights": false,
167
81
  "limit": {
168
- "context": 1048576,
169
- "output": 65536
82
+ "context": 8192,
83
+ "output": 16384
170
84
  },
171
85
  "cost": {
172
- "input": 0.3,
173
- "output": 2.5,
174
- "cache_read": 0.075,
175
- "input_audio": 1
86
+ "input": 0.5,
87
+ "output": 10
176
88
  }
177
89
  },
178
90
  "gemini-2.5-pro": {
179
91
  "id": "gemini-2.5-pro",
180
92
  "name": "Gemini 2.5 Pro",
93
+ "description": "Google's proven reasoning model for coding, math, and multimodal analysis",
181
94
  "family": "gemini-pro",
182
95
  "attachment": true,
183
96
  "reasoning": true,
97
+ "reasoning_options": [
98
+ {
99
+ "type": "budget_tokens",
100
+ "min": 128,
101
+ "max": 32768
102
+ }
103
+ ],
184
104
  "tool_call": true,
185
105
  "structured_output": true,
186
106
  "temperature": true,
187
107
  "knowledge": "2025-01",
188
- "release_date": "2025-03-20",
189
- "last_updated": "2025-06-05",
108
+ "release_date": "2025-06-17",
109
+ "last_updated": "2025-06-17",
190
110
  "modalities": {
191
111
  "input": [
192
112
  "text",
@@ -226,17 +146,29 @@
226
146
  }
227
147
  }
228
148
  },
229
- "gemini-2.5-flash-preview-04-17": {
230
- "id": "gemini-2.5-flash-preview-04-17",
231
- "name": "Gemini 2.5 Flash Preview 04-17",
149
+ "gemini-2.5-flash": {
150
+ "id": "gemini-2.5-flash",
151
+ "name": "Gemini 2.5 Flash",
152
+ "description": "Fast Gemini workhorse for multimodal apps where latency and price matter",
232
153
  "family": "gemini-flash",
233
154
  "attachment": true,
234
155
  "reasoning": true,
156
+ "reasoning_options": [
157
+ {
158
+ "type": "toggle"
159
+ },
160
+ {
161
+ "type": "budget_tokens",
162
+ "min": 0,
163
+ "max": 24576
164
+ }
165
+ ],
235
166
  "tool_call": true,
167
+ "structured_output": true,
236
168
  "temperature": true,
237
169
  "knowledge": "2025-01",
238
- "release_date": "2025-04-17",
239
- "last_updated": "2025-04-17",
170
+ "release_date": "2025-06-17",
171
+ "last_updated": "2025-06-17",
240
172
  "modalities": {
241
173
  "input": [
242
174
  "text",
@@ -255,48 +187,72 @@
255
187
  "output": 65536
256
188
  },
257
189
  "cost": {
258
- "input": 0.15,
259
- "output": 0.6,
260
- "cache_read": 0.0375
190
+ "input": 0.3,
191
+ "output": 2.5,
192
+ "cache_read": 0.03,
193
+ "input_audio": 1
261
194
  }
262
195
  },
263
- "gemma-3-27b-it": {
264
- "id": "gemma-3-27b-it",
265
- "name": "Gemma 3 27B",
266
- "family": "gemma",
196
+ "gemini-3.5-flash": {
197
+ "id": "gemini-3.5-flash",
198
+ "name": "Gemini 3.5 Flash",
199
+ "description": "Fast Gemini model balancing multimodal reasoning, tool use, and cost",
200
+ "family": "gemini-flash",
267
201
  "attachment": true,
268
- "reasoning": false,
202
+ "reasoning": true,
203
+ "reasoning_options": [
204
+ {
205
+ "type": "effort",
206
+ "values": [
207
+ "minimal",
208
+ "low",
209
+ "medium",
210
+ "high"
211
+ ]
212
+ }
213
+ ],
269
214
  "tool_call": true,
270
215
  "structured_output": true,
271
216
  "temperature": true,
272
- "knowledge": "2024-10",
273
- "release_date": "2025-03-12",
274
- "last_updated": "2025-03-12",
217
+ "knowledge": "2025-01",
218
+ "release_date": "2026-05-19",
219
+ "last_updated": "2026-05-19",
275
220
  "modalities": {
276
221
  "input": [
277
222
  "text",
278
- "image"
223
+ "image",
224
+ "video",
225
+ "audio",
226
+ "pdf"
279
227
  ],
280
228
  "output": [
281
229
  "text"
282
230
  ]
283
231
  },
284
- "open_weights": true,
232
+ "open_weights": false,
285
233
  "limit": {
286
- "context": 131072,
287
- "output": 8192
234
+ "context": 1048576,
235
+ "output": 65536
288
236
  },
289
237
  "cost": {
290
- "input": 0,
291
- "output": 0
238
+ "input": 1.5,
239
+ "output": 9,
240
+ "cache_read": 0.15,
241
+ "input_audio": 1.5
292
242
  }
293
243
  },
294
244
  "gemma-4-31b-it": {
295
245
  "id": "gemma-4-31b-it",
296
- "name": "Gemma 4 31B",
246
+ "name": "Gemma 4 31B IT",
247
+ "description": "Largest Gemma 4 instruction model for open, self-hosted chat and reasoning",
297
248
  "family": "gemma",
298
- "attachment": false,
249
+ "attachment": true,
299
250
  "reasoning": true,
251
+ "reasoning_options": [
252
+ {
253
+ "type": "toggle"
254
+ }
255
+ ],
300
256
  "tool_call": true,
301
257
  "structured_output": true,
302
258
  "temperature": true,
@@ -313,57 +269,106 @@
313
269
  },
314
270
  "open_weights": true,
315
271
  "limit": {
316
- "context": 256000,
317
- "output": 8192
272
+ "context": 262144,
273
+ "output": 32768
318
274
  }
319
275
  },
320
- "gemini-2.5-pro-preview-tts": {
321
- "id": "gemini-2.5-pro-preview-tts",
322
- "name": "Gemini 2.5 Pro Preview TTS",
276
+ "gemini-2.0-flash": {
277
+ "id": "gemini-2.0-flash",
278
+ "name": "Gemini 2.0 Flash",
279
+ "description": "Earlier Gemini Flash workhorse for responsive multimodal apps and tool use",
323
280
  "family": "gemini-flash",
281
+ "attachment": true,
282
+ "reasoning": false,
283
+ "tool_call": true,
284
+ "structured_output": true,
285
+ "temperature": true,
286
+ "knowledge": "2024-06",
287
+ "release_date": "2024-12-11",
288
+ "last_updated": "2024-12-11",
289
+ "modalities": {
290
+ "input": [
291
+ "text",
292
+ "image",
293
+ "audio",
294
+ "video",
295
+ "pdf"
296
+ ],
297
+ "output": [
298
+ "text"
299
+ ]
300
+ },
301
+ "open_weights": false,
302
+ "limit": {
303
+ "context": 1048576,
304
+ "output": 8192
305
+ },
306
+ "status": "deprecated",
307
+ "cost": {
308
+ "input": 0.1,
309
+ "output": 0.4,
310
+ "cache_read": 0.025
311
+ }
312
+ },
313
+ "gemini-embedding-001": {
314
+ "id": "gemini-embedding-001",
315
+ "name": "Gemini Embedding 001",
316
+ "description": "Embedding model for semantic search, retrieval, clustering, and ranking pipelines",
317
+ "family": "gemini",
324
318
  "attachment": false,
325
319
  "reasoning": false,
326
320
  "tool_call": false,
327
321
  "temperature": false,
328
- "knowledge": "2025-01",
329
- "release_date": "2025-05-01",
330
- "last_updated": "2025-05-01",
322
+ "knowledge": "2025-05",
323
+ "release_date": "2025-05-20",
324
+ "last_updated": "2025-05-20",
331
325
  "modalities": {
332
326
  "input": [
333
327
  "text"
334
328
  ],
335
329
  "output": [
336
- "audio"
330
+ "text"
337
331
  ]
338
332
  },
339
333
  "open_weights": false,
340
334
  "limit": {
341
- "context": 8000,
342
- "output": 16000
335
+ "context": 2048,
336
+ "output": 1
343
337
  },
344
338
  "cost": {
345
- "input": 1,
346
- "output": 20
339
+ "input": 0.15,
340
+ "output": 0
347
341
  }
348
342
  },
349
- "gemini-2.5-flash-preview-05-20": {
350
- "id": "gemini-2.5-flash-preview-05-20",
351
- "name": "Gemini 2.5 Flash Preview 05-20",
352
- "family": "gemini-flash",
343
+ "gemini-3.1-pro-preview-customtools": {
344
+ "id": "gemini-3.1-pro-preview-customtools",
345
+ "name": "Gemini 3.1 Pro Preview Custom Tools",
346
+ "description": "Advanced Gemini model for complex reasoning, coding, and multimodal analysis",
347
+ "family": "gemini-pro",
353
348
  "attachment": true,
354
349
  "reasoning": true,
350
+ "reasoning_options": [
351
+ {
352
+ "type": "effort",
353
+ "values": [
354
+ "low",
355
+ "medium",
356
+ "high"
357
+ ]
358
+ }
359
+ ],
355
360
  "tool_call": true,
356
361
  "structured_output": true,
357
362
  "temperature": true,
358
363
  "knowledge": "2025-01",
359
- "release_date": "2025-05-20",
360
- "last_updated": "2025-05-20",
364
+ "release_date": "2026-02-19",
365
+ "last_updated": "2026-02-19",
361
366
  "modalities": {
362
367
  "input": [
363
368
  "text",
364
369
  "image",
365
- "audio",
366
370
  "video",
371
+ "audio",
367
372
  "pdf"
368
373
  ],
369
374
  "output": [
@@ -376,95 +381,56 @@
376
381
  "output": 65536
377
382
  },
378
383
  "cost": {
379
- "input": 0.15,
380
- "output": 0.6,
381
- "cache_read": 0.0375
384
+ "input": 2,
385
+ "output": 12,
386
+ "cache_read": 0.2,
387
+ "tiers": [
388
+ {
389
+ "input": 4,
390
+ "output": 18,
391
+ "cache_read": 0.4,
392
+ "tier": {
393
+ "type": "context",
394
+ "size": 200000
395
+ }
396
+ }
397
+ ],
398
+ "context_over_200k": {
399
+ "input": 4,
400
+ "output": 18,
401
+ "cache_read": 0.4
402
+ }
382
403
  }
383
404
  },
384
- "gemini-3.1-flash-image-preview": {
385
- "id": "gemini-3.1-flash-image-preview",
386
- "name": "Gemini 3.1 Flash Image (Preview)",
387
- "family": "gemini-flash",
405
+ "gemini-flash-lite-latest": {
406
+ "id": "gemini-flash-lite-latest",
407
+ "name": "Gemini Flash-Lite Latest",
408
+ "description": "Low-latency Gemini model for high-volume multimodal and agent workloads",
409
+ "family": "gemini-flash-lite",
388
410
  "attachment": true,
389
411
  "reasoning": true,
390
- "tool_call": false,
412
+ "reasoning_options": [
413
+ {
414
+ "type": "toggle"
415
+ },
416
+ {
417
+ "type": "budget_tokens",
418
+ "min": 512,
419
+ "max": 24576
420
+ }
421
+ ],
422
+ "tool_call": true,
423
+ "structured_output": true,
391
424
  "temperature": true,
392
425
  "knowledge": "2025-01",
393
- "release_date": "2026-02-26",
394
- "last_updated": "2026-02-26",
426
+ "release_date": "2025-09-25",
427
+ "last_updated": "2025-09-25",
395
428
  "modalities": {
396
429
  "input": [
397
430
  "text",
398
431
  "image",
399
- "pdf"
400
- ],
401
- "output": [
402
- "text",
403
- "image"
404
- ]
405
- },
406
- "open_weights": false,
407
- "limit": {
408
- "context": 131072,
409
- "output": 32768
410
- },
411
- "cost": {
412
- "input": 0.5,
413
- "output": 60
414
- }
415
- },
416
- "gemini-live-2.5-flash-preview-native-audio": {
417
- "id": "gemini-live-2.5-flash-preview-native-audio",
418
- "name": "Gemini Live 2.5 Flash Preview Native Audio",
419
- "family": "gemini-flash",
420
- "attachment": false,
421
- "reasoning": true,
422
- "tool_call": true,
423
- "temperature": false,
424
- "knowledge": "2025-01",
425
- "release_date": "2025-06-17",
426
- "last_updated": "2025-09-18",
427
- "modalities": {
428
- "input": [
429
- "text",
430
- "audio",
431
- "video"
432
- ],
433
- "output": [
434
- "text",
435
- "audio"
436
- ]
437
- },
438
- "open_weights": false,
439
- "limit": {
440
- "context": 131072,
441
- "output": 65536
442
- },
443
- "cost": {
444
- "input": 0.5,
445
- "output": 2,
446
- "input_audio": 3,
447
- "output_audio": 12
448
- }
449
- },
450
- "gemini-2.5-pro-preview-06-05": {
451
- "id": "gemini-2.5-pro-preview-06-05",
452
- "name": "Gemini 2.5 Pro Preview 06-05",
453
- "family": "gemini-pro",
454
- "attachment": true,
455
- "reasoning": true,
456
- "tool_call": true,
457
- "structured_output": true,
458
- "temperature": true,
459
- "knowledge": "2025-01",
460
- "release_date": "2025-06-05",
461
- "last_updated": "2025-06-05",
462
- "modalities": {
463
- "input": [
464
- "text",
465
- "image",
466
- "audio",
467
- "video",
432
+ "audio",
433
+ "video",
468
434
  "pdf"
469
435
  ],
470
436
  "output": [
@@ -477,145 +443,95 @@
477
443
  "output": 65536
478
444
  },
479
445
  "cost": {
480
- "input": 1.25,
481
- "output": 10,
482
- "cache_read": 0.31
483
- }
484
- },
485
- "gemma-3n-e2b-it": {
486
- "id": "gemma-3n-e2b-it",
487
- "name": "Gemma 3n 2B",
488
- "family": "gemma",
489
- "attachment": true,
490
- "reasoning": false,
491
- "tool_call": false,
492
- "temperature": true,
493
- "knowledge": "2024-10",
494
- "release_date": "2025-07-09",
495
- "last_updated": "2025-07-09",
496
- "modalities": {
497
- "input": [
498
- "text"
499
- ],
500
- "output": [
501
- "text"
502
- ]
503
- },
504
- "open_weights": true,
505
- "limit": {
506
- "context": 8192,
507
- "output": 2000
508
- },
509
- "cost": {
510
- "input": 0,
511
- "output": 0
512
- }
513
- },
514
- "gemma-3n-e4b-it": {
515
- "id": "gemma-3n-e4b-it",
516
- "name": "Gemma 3n 4B",
517
- "family": "gemma",
518
- "attachment": true,
519
- "reasoning": false,
520
- "tool_call": false,
521
- "temperature": true,
522
- "knowledge": "2024-10",
523
- "release_date": "2025-05-20",
524
- "last_updated": "2025-05-20",
525
- "modalities": {
526
- "input": [
527
- "text"
528
- ],
529
- "output": [
530
- "text"
531
- ]
532
- },
533
- "open_weights": true,
534
- "limit": {
535
- "context": 8192,
536
- "output": 2000
537
- },
538
- "cost": {
539
- "input": 0,
540
- "output": 0
446
+ "input": 0.1,
447
+ "output": 0.4,
448
+ "cache_read": 0.025
541
449
  }
542
450
  },
543
- "gemini-2.5-pro-preview-05-06": {
544
- "id": "gemini-2.5-pro-preview-05-06",
545
- "name": "Gemini 2.5 Pro Preview 05-06",
451
+ "gemini-3-pro-image-preview": {
452
+ "id": "gemini-3-pro-image-preview",
453
+ "name": "Nano Banana Pro",
454
+ "description": "Nano Banana Pro for higher-fidelity image generation and design-heavy edits",
546
455
  "family": "gemini-pro",
547
456
  "attachment": true,
548
457
  "reasoning": true,
549
- "tool_call": true,
550
- "structured_output": true,
458
+ "reasoning_options": [],
459
+ "tool_call": false,
551
460
  "temperature": true,
552
461
  "knowledge": "2025-01",
553
- "release_date": "2025-05-06",
554
- "last_updated": "2025-05-06",
462
+ "release_date": "2025-11-20",
463
+ "last_updated": "2025-11-20",
555
464
  "modalities": {
556
465
  "input": [
557
466
  "text",
558
- "image",
559
- "audio",
560
- "video",
561
- "pdf"
467
+ "image"
562
468
  ],
563
469
  "output": [
564
- "text"
470
+ "text",
471
+ "image"
565
472
  ]
566
473
  },
567
474
  "open_weights": false,
568
475
  "limit": {
569
- "context": 1048576,
570
- "output": 65536
476
+ "context": 131072,
477
+ "output": 32768
571
478
  },
572
479
  "cost": {
573
- "input": 1.25,
574
- "output": 10,
575
- "cache_read": 0.31
480
+ "input": 2,
481
+ "output": 120
576
482
  }
577
483
  },
578
- "gemini-2.5-flash-lite-preview-09-2025": {
579
- "id": "gemini-2.5-flash-lite-preview-09-2025",
580
- "name": "Gemini 2.5 Flash Lite Preview 09-25",
581
- "family": "gemini-flash-lite",
484
+ "gemini-2.5-flash-image": {
485
+ "id": "gemini-2.5-flash-image",
486
+ "name": "Nano Banana",
487
+ "description": "Nano Banana image model for fast generation, edits, and character-consistent assets",
488
+ "family": "gemini-flash",
582
489
  "attachment": true,
583
490
  "reasoning": true,
584
- "tool_call": true,
585
- "structured_output": true,
491
+ "reasoning_options": [],
492
+ "tool_call": false,
586
493
  "temperature": true,
587
- "knowledge": "2025-01",
588
- "release_date": "2025-09-25",
589
- "last_updated": "2025-09-25",
494
+ "knowledge": "2025-06",
495
+ "release_date": "2025-08-26",
496
+ "last_updated": "2025-08-26",
590
497
  "modalities": {
591
498
  "input": [
592
499
  "text",
593
- "image",
594
- "audio",
595
- "video",
596
- "pdf"
500
+ "image"
597
501
  ],
598
502
  "output": [
599
- "text"
503
+ "text",
504
+ "image"
600
505
  ]
601
506
  },
602
507
  "open_weights": false,
603
508
  "limit": {
604
- "context": 1048576,
605
- "output": 65536
509
+ "context": 32768,
510
+ "output": 32768
606
511
  },
607
512
  "cost": {
608
- "input": 0.1,
609
- "output": 0.4,
610
- "cache_read": 0.025
513
+ "input": 0.3,
514
+ "output": 30,
515
+ "cache_read": 0.075
611
516
  }
612
517
  },
613
518
  "gemini-2.5-flash-lite": {
614
519
  "id": "gemini-2.5-flash-lite",
615
- "name": "Gemini 2.5 Flash Lite",
520
+ "name": "Gemini 2.5 Flash-Lite",
521
+ "description": "Lean Gemini 2.5 lane for cheap multimodal traffic and quick agents",
616
522
  "family": "gemini-flash-lite",
617
523
  "attachment": true,
618
524
  "reasoning": true,
525
+ "reasoning_options": [
526
+ {
527
+ "type": "toggle"
528
+ },
529
+ {
530
+ "type": "budget_tokens",
531
+ "min": 512,
532
+ "max": 24576
533
+ }
534
+ ],
619
535
  "tool_call": true,
620
536
  "structured_output": true,
621
537
  "temperature": true,
@@ -646,53 +562,71 @@
646
562
  "input_audio": 0.3
647
563
  }
648
564
  },
649
- "gemini-flash-lite-latest": {
650
- "id": "gemini-flash-lite-latest",
651
- "name": "Gemini Flash-Lite Latest",
652
- "family": "gemini-flash-lite",
565
+ "gemini-3.1-flash-image-preview": {
566
+ "id": "gemini-3.1-flash-image-preview",
567
+ "name": "Nano Banana 2",
568
+ "description": "Image model for prompt-driven generation, editing, and visual design workflows",
569
+ "family": "gemini-flash",
653
570
  "attachment": true,
654
571
  "reasoning": true,
655
- "tool_call": true,
656
- "structured_output": true,
572
+ "reasoning_options": [
573
+ {
574
+ "type": "effort",
575
+ "values": [
576
+ "minimal",
577
+ "high"
578
+ ]
579
+ }
580
+ ],
581
+ "tool_call": false,
657
582
  "temperature": true,
658
583
  "knowledge": "2025-01",
659
- "release_date": "2025-09-25",
660
- "last_updated": "2025-09-25",
584
+ "release_date": "2026-02-26",
585
+ "last_updated": "2026-02-26",
661
586
  "modalities": {
662
587
  "input": [
663
588
  "text",
664
589
  "image",
665
- "audio",
666
- "video",
667
590
  "pdf"
668
591
  ],
669
592
  "output": [
670
- "text"
593
+ "text",
594
+ "image"
671
595
  ]
672
596
  },
673
597
  "open_weights": false,
674
598
  "limit": {
675
- "context": 1048576,
599
+ "context": 65536,
676
600
  "output": 65536
677
601
  },
678
602
  "cost": {
679
- "input": 0.1,
680
- "output": 0.4,
681
- "cache_read": 0.025
603
+ "input": 0.5,
604
+ "output": 60
682
605
  }
683
606
  },
684
- "gemini-3-pro-preview": {
685
- "id": "gemini-3-pro-preview",
686
- "name": "Gemini 3 Pro Preview",
607
+ "gemini-3.1-pro-preview": {
608
+ "id": "gemini-3.1-pro-preview",
609
+ "name": "Gemini 3.1 Pro Preview",
610
+ "description": "Reasoning-first Gemini preview for agentic coding and complex problem solving",
687
611
  "family": "gemini-pro",
688
612
  "attachment": true,
689
613
  "reasoning": true,
614
+ "reasoning_options": [
615
+ {
616
+ "type": "effort",
617
+ "values": [
618
+ "low",
619
+ "medium",
620
+ "high"
621
+ ]
622
+ }
623
+ ],
690
624
  "tool_call": true,
691
625
  "structured_output": true,
692
626
  "temperature": true,
693
627
  "knowledge": "2025-01",
694
- "release_date": "2025-11-18",
695
- "last_updated": "2025-11-18",
628
+ "release_date": "2026-02-19",
629
+ "last_updated": "2026-02-19",
696
630
  "modalities": {
697
631
  "input": [
698
632
  "text",
@@ -707,8 +641,8 @@
707
641
  },
708
642
  "open_weights": false,
709
643
  "limit": {
710
- "context": 1000000,
711
- "output": 64000
644
+ "context": 1048576,
645
+ "output": 65536
712
646
  },
713
647
  "cost": {
714
648
  "input": 2,
@@ -732,52 +666,60 @@
732
666
  }
733
667
  }
734
668
  },
735
- "gemini-2.0-flash-lite": {
736
- "id": "gemini-2.0-flash-lite",
737
- "name": "Gemini 2.0 Flash Lite",
738
- "family": "gemini-flash-lite",
669
+ "gemma-4-26b-a4b-it": {
670
+ "id": "gemma-4-26b-a4b-it",
671
+ "name": "Gemma 4 26B A4B IT",
672
+ "description": "Open Gemma instruction model for efficient chat and self-hosted deployments",
673
+ "family": "gemma",
739
674
  "attachment": true,
740
- "reasoning": false,
675
+ "reasoning": true,
676
+ "reasoning_options": [
677
+ {
678
+ "type": "toggle"
679
+ }
680
+ ],
741
681
  "tool_call": true,
742
682
  "structured_output": true,
743
683
  "temperature": true,
744
- "knowledge": "2024-06",
745
- "release_date": "2024-12-11",
746
- "last_updated": "2024-12-11",
684
+ "release_date": "2026-04-02",
685
+ "last_updated": "2026-04-02",
747
686
  "modalities": {
748
687
  "input": [
749
688
  "text",
750
- "image",
751
- "audio",
752
- "video",
753
- "pdf"
689
+ "image"
754
690
  ],
755
691
  "output": [
756
692
  "text"
757
693
  ]
758
694
  },
759
- "open_weights": false,
695
+ "open_weights": true,
760
696
  "limit": {
761
- "context": 1048576,
762
- "output": 8192
763
- },
764
- "cost": {
765
- "input": 0.075,
766
- "output": 0.3
697
+ "context": 262144,
698
+ "output": 32768
767
699
  }
768
700
  },
769
- "gemini-3.1-pro-preview-customtools": {
770
- "id": "gemini-3.1-pro-preview-customtools",
771
- "name": "Gemini 3.1 Pro Preview Custom Tools",
701
+ "gemini-3-pro-preview": {
702
+ "id": "gemini-3-pro-preview",
703
+ "name": "Gemini 3 Pro Preview",
704
+ "description": "Preview Gemini flagship for complex reasoning, coding, and rich multimodal prompts",
772
705
  "family": "gemini-pro",
773
706
  "attachment": true,
774
707
  "reasoning": true,
708
+ "reasoning_options": [
709
+ {
710
+ "type": "effort",
711
+ "values": [
712
+ "low",
713
+ "high"
714
+ ]
715
+ }
716
+ ],
775
717
  "tool_call": true,
776
718
  "structured_output": true,
777
719
  "temperature": true,
778
720
  "knowledge": "2025-01",
779
- "release_date": "2026-02-19",
780
- "last_updated": "2026-02-19",
721
+ "release_date": "2025-11-18",
722
+ "last_updated": "2025-11-18",
781
723
  "modalities": {
782
724
  "input": [
783
725
  "text",
@@ -795,6 +737,7 @@
795
737
  "context": 1048576,
796
738
  "output": 65536
797
739
  },
740
+ "status": "deprecated",
798
741
  "cost": {
799
742
  "input": 2,
800
743
  "output": 12,
@@ -817,46 +760,63 @@
817
760
  }
818
761
  }
819
762
  },
820
- "gemini-2.5-flash-image-preview": {
821
- "id": "gemini-2.5-flash-image-preview",
822
- "name": "Gemini 2.5 Flash Image (Preview)",
763
+ "gemini-3-flash-preview": {
764
+ "id": "gemini-3-flash-preview",
765
+ "name": "Gemini 3 Flash Preview",
766
+ "description": "New Gemini flash lane bringing frontier-style multimodal reasoning to cheaper runs",
823
767
  "family": "gemini-flash",
824
768
  "attachment": true,
825
769
  "reasoning": true,
826
- "tool_call": false,
770
+ "reasoning_options": [
771
+ {
772
+ "type": "effort",
773
+ "values": [
774
+ "minimal",
775
+ "low",
776
+ "medium",
777
+ "high"
778
+ ]
779
+ }
780
+ ],
781
+ "tool_call": true,
782
+ "structured_output": true,
827
783
  "temperature": true,
828
- "knowledge": "2025-06",
829
- "release_date": "2025-08-26",
830
- "last_updated": "2025-08-26",
784
+ "knowledge": "2025-01",
785
+ "release_date": "2025-12-17",
786
+ "last_updated": "2025-12-17",
831
787
  "modalities": {
832
788
  "input": [
833
789
  "text",
834
- "image"
790
+ "image",
791
+ "video",
792
+ "audio",
793
+ "pdf"
835
794
  ],
836
795
  "output": [
837
- "text",
838
- "image"
796
+ "text"
839
797
  ]
840
798
  },
841
799
  "open_weights": false,
842
800
  "limit": {
843
- "context": 32768,
844
- "output": 32768
801
+ "context": 1048576,
802
+ "output": 65536
845
803
  },
846
804
  "cost": {
847
- "input": 0.3,
848
- "output": 30,
849
- "cache_read": 0.075
805
+ "input": 0.5,
806
+ "output": 3,
807
+ "cache_read": 0.05,
808
+ "input_audio": 1
850
809
  }
851
810
  },
852
- "gemini-2.5-flash-preview-tts": {
853
- "id": "gemini-2.5-flash-preview-tts",
854
- "name": "Gemini 2.5 Flash Preview TTS",
811
+ "gemini-2.5-pro-preview-tts": {
812
+ "id": "gemini-2.5-pro-preview-tts",
813
+ "name": "Gemini 2.5 Pro Preview TTS",
814
+ "description": "Speech generation model for controllable voice, narration, and audio delivery",
855
815
  "family": "gemini-flash",
856
816
  "attachment": false,
857
817
  "reasoning": false,
858
818
  "tool_call": false,
859
- "temperature": false,
819
+ "temperature": true,
860
820
  "knowledge": "2025-01",
861
821
  "release_date": "2025-05-01",
862
822
  "last_updated": "2025-05-01",
@@ -870,25 +830,37 @@
870
830
  },
871
831
  "open_weights": false,
872
832
  "limit": {
873
- "context": 8000,
874
- "output": 16000
833
+ "context": 8192,
834
+ "output": 16384
875
835
  },
876
836
  "cost": {
877
- "input": 0.5,
878
- "output": 10
837
+ "input": 1,
838
+ "output": 20
879
839
  }
880
840
  },
881
- "gemini-2.5-flash-lite-preview-06-17": {
882
- "id": "gemini-2.5-flash-lite-preview-06-17",
883
- "name": "Gemini 2.5 Flash Lite Preview 06-17",
884
- "family": "gemini-flash-lite",
841
+ "gemini-flash-latest": {
842
+ "id": "gemini-flash-latest",
843
+ "name": "Gemini Flash Latest",
844
+ "description": "Fast Gemini model balancing multimodal reasoning, tool use, and cost",
845
+ "family": "gemini-flash",
885
846
  "attachment": true,
886
847
  "reasoning": true,
848
+ "reasoning_options": [
849
+ {
850
+ "type": "toggle"
851
+ },
852
+ {
853
+ "type": "budget_tokens",
854
+ "min": 0,
855
+ "max": 24576
856
+ }
857
+ ],
887
858
  "tool_call": true,
859
+ "structured_output": true,
888
860
  "temperature": true,
889
861
  "knowledge": "2025-01",
890
- "release_date": "2025-06-17",
891
- "last_updated": "2025-06-17",
862
+ "release_date": "2025-09-25",
863
+ "last_updated": "2025-09-25",
892
864
  "modalities": {
893
865
  "input": [
894
866
  "text",
@@ -907,18 +879,30 @@
907
879
  "output": 65536
908
880
  },
909
881
  "cost": {
910
- "input": 0.1,
911
- "output": 0.4,
912
- "cache_read": 0.025,
913
- "input_audio": 0.3
882
+ "input": 0.3,
883
+ "output": 2.5,
884
+ "cache_read": 0.075,
885
+ "input_audio": 1
914
886
  }
915
887
  },
916
888
  "gemini-3.1-flash-lite-preview": {
917
889
  "id": "gemini-3.1-flash-lite-preview",
918
890
  "name": "Gemini 3.1 Flash Lite Preview",
891
+ "description": "Legacy model retained for compatibility with older integrations",
919
892
  "family": "gemini-flash-lite",
920
893
  "attachment": true,
921
894
  "reasoning": true,
895
+ "reasoning_options": [
896
+ {
897
+ "type": "effort",
898
+ "values": [
899
+ "minimal",
900
+ "low",
901
+ "medium",
902
+ "high"
903
+ ]
904
+ }
905
+ ],
922
906
  "tool_call": true,
923
907
  "structured_output": true,
924
908
  "temperature": true,
@@ -942,6 +926,7 @@
942
926
  "context": 1048576,
943
927
  "output": 65536
944
928
  },
929
+ "status": "deprecated",
945
930
  "cost": {
946
931
  "input": 0.25,
947
932
  "output": 1.5,
@@ -949,231 +934,11 @@
949
934
  "input_audio": 0.5
950
935
  }
951
936
  },
952
- "gemini-3.1-pro-preview": {
953
- "id": "gemini-3.1-pro-preview",
954
- "name": "Gemini 3.1 Pro Preview",
955
- "family": "gemini-pro",
956
- "attachment": true,
957
- "reasoning": true,
958
- "tool_call": true,
959
- "structured_output": true,
960
- "temperature": true,
961
- "knowledge": "2025-01",
962
- "release_date": "2026-02-19",
963
- "last_updated": "2026-02-19",
964
- "modalities": {
965
- "input": [
966
- "text",
967
- "image",
968
- "video",
969
- "audio",
970
- "pdf"
971
- ],
972
- "output": [
973
- "text"
974
- ]
975
- },
976
- "open_weights": false,
977
- "limit": {
978
- "context": 1048576,
979
- "output": 65536
980
- },
981
- "cost": {
982
- "input": 2,
983
- "output": 12,
984
- "cache_read": 0.2,
985
- "tiers": [
986
- {
987
- "input": 4,
988
- "output": 18,
989
- "cache_read": 0.4,
990
- "tier": {
991
- "type": "context",
992
- "size": 200000
993
- }
994
- }
995
- ],
996
- "context_over_200k": {
997
- "input": 4,
998
- "output": 18,
999
- "cache_read": 0.4
1000
- }
1001
- }
1002
- },
1003
- "gemini-3-flash-preview": {
1004
- "id": "gemini-3-flash-preview",
1005
- "name": "Gemini 3 Flash Preview",
1006
- "family": "gemini-flash",
1007
- "attachment": true,
1008
- "reasoning": true,
1009
- "tool_call": true,
1010
- "structured_output": true,
1011
- "temperature": true,
1012
- "knowledge": "2025-01",
1013
- "release_date": "2025-12-17",
1014
- "last_updated": "2025-12-17",
1015
- "modalities": {
1016
- "input": [
1017
- "text",
1018
- "image",
1019
- "video",
1020
- "audio",
1021
- "pdf"
1022
- ],
1023
- "output": [
1024
- "text"
1025
- ]
1026
- },
1027
- "open_weights": false,
1028
- "limit": {
1029
- "context": 1048576,
1030
- "output": 65536
1031
- },
1032
- "cost": {
1033
- "input": 0.5,
1034
- "output": 3,
1035
- "cache_read": 0.05,
1036
- "input_audio": 1
1037
- }
1038
- },
1039
- "gemini-live-2.5-flash": {
1040
- "id": "gemini-live-2.5-flash",
1041
- "name": "Gemini Live 2.5 Flash",
1042
- "family": "gemini-flash",
1043
- "attachment": true,
1044
- "reasoning": true,
1045
- "tool_call": true,
1046
- "temperature": true,
1047
- "knowledge": "2025-01",
1048
- "release_date": "2025-09-01",
1049
- "last_updated": "2025-09-01",
1050
- "modalities": {
1051
- "input": [
1052
- "text",
1053
- "image",
1054
- "audio",
1055
- "video"
1056
- ],
1057
- "output": [
1058
- "text",
1059
- "audio"
1060
- ]
1061
- },
1062
- "open_weights": false,
1063
- "limit": {
1064
- "context": 128000,
1065
- "output": 8000
1066
- },
1067
- "cost": {
1068
- "input": 0.5,
1069
- "output": 2,
1070
- "input_audio": 3,
1071
- "output_audio": 12
1072
- }
1073
- },
1074
- "gemma-3-12b-it": {
1075
- "id": "gemma-3-12b-it",
1076
- "name": "Gemma 3 12B",
1077
- "family": "gemma",
1078
- "attachment": true,
1079
- "reasoning": false,
1080
- "tool_call": false,
1081
- "structured_output": true,
1082
- "temperature": true,
1083
- "knowledge": "2024-10",
1084
- "release_date": "2025-03-13",
1085
- "last_updated": "2025-03-13",
1086
- "modalities": {
1087
- "input": [
1088
- "text",
1089
- "image"
1090
- ],
1091
- "output": [
1092
- "text"
1093
- ]
1094
- },
1095
- "open_weights": true,
1096
- "limit": {
1097
- "context": 32768,
1098
- "output": 8192
1099
- },
1100
- "cost": {
1101
- "input": 0,
1102
- "output": 0
1103
- }
1104
- },
1105
- "gemini-2.5-flash": {
1106
- "id": "gemini-2.5-flash",
1107
- "name": "Gemini 2.5 Flash",
1108
- "family": "gemini-flash",
1109
- "attachment": true,
1110
- "reasoning": true,
1111
- "tool_call": true,
1112
- "structured_output": true,
1113
- "temperature": true,
1114
- "knowledge": "2025-01",
1115
- "release_date": "2025-03-20",
1116
- "last_updated": "2025-06-05",
1117
- "modalities": {
1118
- "input": [
1119
- "text",
1120
- "image",
1121
- "audio",
1122
- "video",
1123
- "pdf"
1124
- ],
1125
- "output": [
1126
- "text"
1127
- ]
1128
- },
1129
- "open_weights": false,
1130
- "limit": {
1131
- "context": 1048576,
1132
- "output": 65536
1133
- },
1134
- "cost": {
1135
- "input": 0.3,
1136
- "output": 2.5,
1137
- "cache_read": 0.03,
1138
- "input_audio": 1
1139
- }
1140
- },
1141
- "gemini-2.5-flash-image": {
1142
- "id": "gemini-2.5-flash-image",
1143
- "name": "Gemini 2.5 Flash Image",
1144
- "family": "gemini-flash",
1145
- "attachment": true,
1146
- "reasoning": true,
1147
- "tool_call": false,
1148
- "temperature": true,
1149
- "knowledge": "2025-06",
1150
- "release_date": "2025-08-26",
1151
- "last_updated": "2025-08-26",
1152
- "modalities": {
1153
- "input": [
1154
- "text",
1155
- "image"
1156
- ],
1157
- "output": [
1158
- "text",
1159
- "image"
1160
- ]
1161
- },
1162
- "open_weights": false,
1163
- "limit": {
1164
- "context": 32768,
1165
- "output": 32768
1166
- },
1167
- "cost": {
1168
- "input": 0.3,
1169
- "output": 30,
1170
- "cache_read": 0.075
1171
- }
1172
- },
1173
- "gemini-2.0-flash": {
1174
- "id": "gemini-2.0-flash",
1175
- "name": "Gemini 2.0 Flash",
1176
- "family": "gemini-flash",
937
+ "gemini-2.0-flash-lite": {
938
+ "id": "gemini-2.0-flash-lite",
939
+ "name": "Gemini 2.0 Flash-Lite",
940
+ "description": "Legacy model retained for compatibility with older integrations",
941
+ "family": "gemini-flash-lite",
1177
942
  "attachment": true,
1178
943
  "reasoning": false,
1179
944
  "tool_call": true,
@@ -1199,135 +964,10 @@
1199
964
  "context": 1048576,
1200
965
  "output": 8192
1201
966
  },
967
+ "status": "deprecated",
1202
968
  "cost": {
1203
- "input": 0.1,
1204
- "output": 0.4,
1205
- "cache_read": 0.025
1206
- }
1207
- },
1208
- "gemini-1.5-pro": {
1209
- "id": "gemini-1.5-pro",
1210
- "name": "Gemini 1.5 Pro",
1211
- "family": "gemini-pro",
1212
- "attachment": true,
1213
- "reasoning": false,
1214
- "tool_call": true,
1215
- "temperature": true,
1216
- "knowledge": "2024-04",
1217
- "release_date": "2024-02-15",
1218
- "last_updated": "2024-02-15",
1219
- "modalities": {
1220
- "input": [
1221
- "text",
1222
- "image",
1223
- "audio",
1224
- "video"
1225
- ],
1226
- "output": [
1227
- "text"
1228
- ]
1229
- },
1230
- "open_weights": false,
1231
- "limit": {
1232
- "context": 1000000,
1233
- "output": 8192
1234
- },
1235
- "cost": {
1236
- "input": 1.25,
1237
- "output": 5,
1238
- "cache_read": 0.3125
1239
- }
1240
- },
1241
- "gemini-embedding-001": {
1242
- "id": "gemini-embedding-001",
1243
- "name": "Gemini Embedding 001",
1244
- "family": "gemini",
1245
- "attachment": false,
1246
- "reasoning": false,
1247
- "tool_call": false,
1248
- "temperature": false,
1249
- "knowledge": "2025-05",
1250
- "release_date": "2025-05-20",
1251
- "last_updated": "2025-05-20",
1252
- "modalities": {
1253
- "input": [
1254
- "text"
1255
- ],
1256
- "output": [
1257
- "text"
1258
- ]
1259
- },
1260
- "open_weights": false,
1261
- "limit": {
1262
- "context": 2048,
1263
- "output": 3072
1264
- },
1265
- "cost": {
1266
- "input": 0.15,
1267
- "output": 0
1268
- }
1269
- },
1270
- "gemma-3-4b-it": {
1271
- "id": "gemma-3-4b-it",
1272
- "name": "Gemma 3 4B",
1273
- "family": "gemma",
1274
- "attachment": true,
1275
- "reasoning": false,
1276
- "tool_call": false,
1277
- "temperature": true,
1278
- "knowledge": "2024-10",
1279
- "release_date": "2025-03-13",
1280
- "last_updated": "2025-03-13",
1281
- "modalities": {
1282
- "input": [
1283
- "text",
1284
- "image"
1285
- ],
1286
- "output": [
1287
- "text"
1288
- ]
1289
- },
1290
- "open_weights": true,
1291
- "limit": {
1292
- "context": 32768,
1293
- "output": 8192
1294
- },
1295
- "cost": {
1296
- "input": 0,
1297
- "output": 0
1298
- }
1299
- },
1300
- "gemini-1.5-flash-8b": {
1301
- "id": "gemini-1.5-flash-8b",
1302
- "name": "Gemini 1.5 Flash-8B",
1303
- "family": "gemini-flash",
1304
- "attachment": true,
1305
- "reasoning": false,
1306
- "tool_call": true,
1307
- "temperature": true,
1308
- "knowledge": "2024-04",
1309
- "release_date": "2024-10-03",
1310
- "last_updated": "2024-10-03",
1311
- "modalities": {
1312
- "input": [
1313
- "text",
1314
- "image",
1315
- "audio",
1316
- "video"
1317
- ],
1318
- "output": [
1319
- "text"
1320
- ]
1321
- },
1322
- "open_weights": false,
1323
- "limit": {
1324
- "context": 1000000,
1325
- "output": 8192
1326
- },
1327
- "cost": {
1328
- "input": 0.0375,
1329
- "output": 0.15,
1330
- "cache_read": 0.01
969
+ "input": 0.075,
970
+ "output": 0.3
1331
971
  }
1332
972
  }
1333
973
  }