llm.rb 15.2.2 → 15.4.0

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 (79) hide show
  1. checksums.yaml +4 -4
  2. data/CHANGELOG.md +197 -3
  3. data/README.md +180 -64
  4. data/bin/llm.rb +9 -2
  5. data/data/alibaba.json +45 -0
  6. data/data/anthropic.json +67 -0
  7. data/data/bedrock.json +1466 -397
  8. data/data/deepinfra.json +148 -16
  9. data/data/deepseek.json +3 -0
  10. data/data/mistral.json +42 -0
  11. data/data/openai.json +186 -0
  12. data/data/openrouter.json +1538 -378
  13. data/data/xai.json +53 -20
  14. data/data/zai.json +90 -4
  15. data/docs/deepdive/advanced/compaction.md +1 -2
  16. data/docs/deepdive/advanced/context.md +214 -1
  17. data/docs/deepdive/advanced/guard.md +9 -57
  18. data/docs/deepdive/features/builtin_tools.md +14 -16
  19. data/docs/deepdive/features/console.md +5 -0
  20. data/docs/deepdive/features/database.md +85 -10
  21. data/docs/deepdive/fundamentals/agents.md +7 -8
  22. data/docs/deepdive/fundamentals/providers.md +45 -5
  23. data/docs/deepdive/fundamentals/schema.md +73 -0
  24. data/docs/deepdive/fundamentals/tools.md +80 -27
  25. data/docs/deepdive/media/audio.md +8 -19
  26. data/docs/deepdive/media/images.md +8 -10
  27. data/docs/deepdive/media/ocr.md +1 -3
  28. data/docs/deepdive/reference/cost.md +48 -0
  29. data/docs/deepdive/reference/tracer.md +76 -0
  30. data/docs/deepdive.md +1 -1
  31. data/lib/llm/active_record/message.rb +113 -0
  32. data/lib/llm/active_record.rb +1 -0
  33. data/lib/llm/agent.rb +44 -19
  34. data/lib/llm/console/buffer.rb +9 -1
  35. data/lib/llm/console.rb +6 -1
  36. data/lib/llm/context/deserializer.rb +10 -3
  37. data/lib/llm/context.rb +62 -26
  38. data/lib/llm/guard.rb +2 -8
  39. data/lib/llm/message.rb +18 -7
  40. data/lib/llm/provider.rb +74 -16
  41. data/lib/llm/providers/alibaba.rb +15 -0
  42. data/lib/llm/providers/anthropic/error_handler.rb +5 -2
  43. data/lib/llm/providers/anthropic/files.rb +12 -12
  44. data/lib/llm/providers/anthropic/models.rb +2 -2
  45. data/lib/llm/providers/anthropic.rb +5 -3
  46. data/lib/llm/providers/bedrock/error_handler.rb +3 -2
  47. data/lib/llm/providers/bedrock/models.rb +5 -3
  48. data/lib/llm/providers/bedrock.rb +5 -3
  49. data/lib/llm/providers/deepinfra/audio.rb +4 -4
  50. data/lib/llm/providers/deepinfra/images.rb +4 -4
  51. data/lib/llm/providers/google/error_handler.rb +5 -2
  52. data/lib/llm/providers/google/files.rb +10 -10
  53. data/lib/llm/providers/google/images.rb +2 -2
  54. data/lib/llm/providers/google/models.rb +2 -2
  55. data/lib/llm/providers/google.rb +7 -7
  56. data/lib/llm/providers/mistral.rb +3 -1
  57. data/lib/llm/providers/ollama/error_handler.rb +5 -2
  58. data/lib/llm/providers/ollama/models.rb +2 -2
  59. data/lib/llm/providers/ollama.rb +7 -5
  60. data/lib/llm/providers/openai/audio.rb +6 -6
  61. data/lib/llm/providers/openai/error_handler.rb +5 -2
  62. data/lib/llm/providers/openai/files.rb +10 -10
  63. data/lib/llm/providers/openai/images.rb +4 -4
  64. data/lib/llm/providers/openai/models.rb +2 -2
  65. data/lib/llm/providers/openai/moderations.rb +2 -2
  66. data/lib/llm/providers/openai/request_adapter.rb +1 -1
  67. data/lib/llm/providers/openai/responses.rb +8 -8
  68. data/lib/llm/providers/openai/vector_stores.rb +22 -22
  69. data/lib/llm/providers/openai.rb +10 -8
  70. data/lib/llm/providers/xai/images.rb +4 -4
  71. data/lib/llm/schema.rb +24 -0
  72. data/lib/llm/tracer/telemetry.rb +4 -4
  73. data/lib/llm/tracer.rb +11 -3
  74. data/lib/llm/transport/execution.rb +8 -4
  75. data/lib/llm/utils.rb +13 -0
  76. data/lib/llm/version.rb +1 -1
  77. data/llm.gemspec +2 -2
  78. metadata +5 -5
  79. data/lib/llm/guard/loop.rb +0 -89
data/data/deepinfra.json CHANGED
@@ -314,7 +314,7 @@
314
314
  "structured_output": true,
315
315
  "temperature": true,
316
316
  "release_date": "2026-08-21",
317
- "last_updated": "2026-08-21",
317
+ "last_updated": "2026-09-01",
318
318
  "modalities": {
319
319
  "input": [
320
320
  "text",
@@ -324,7 +324,7 @@
324
324
  "text"
325
325
  ]
326
326
  },
327
- "open_weights": false,
327
+ "open_weights": true,
328
328
  "limit": {
329
329
  "context": 1048576,
330
330
  "output": 384000
@@ -483,8 +483,8 @@
483
483
  "output": 384000
484
484
  },
485
485
  "cost": {
486
- "input": 0.3,
487
- "output": 1.2,
486
+ "input": 0.2,
487
+ "output": 0.6,
488
488
  "cache_read": 0.006
489
489
  }
490
490
  },
@@ -721,6 +721,38 @@
721
721
  "cache_read": 0.025
722
722
  }
723
723
  },
724
+ "google/gemma-3-4b-it": {
725
+ "id": "google/gemma-3-4b-it",
726
+ "name": "Gemma 3 4B IT",
727
+ "description": "Open multimodal Gemma instruction model for efficient text generation and image understanding",
728
+ "family": "gemma",
729
+ "attachment": true,
730
+ "reasoning": false,
731
+ "tool_call": true,
732
+ "structured_output": true,
733
+ "temperature": true,
734
+ "knowledge": "2024-08",
735
+ "release_date": "2025-03-12",
736
+ "last_updated": "2025-03-12",
737
+ "modalities": {
738
+ "input": [
739
+ "text",
740
+ "image"
741
+ ],
742
+ "output": [
743
+ "text"
744
+ ]
745
+ },
746
+ "open_weights": true,
747
+ "limit": {
748
+ "context": 131072,
749
+ "output": 131072
750
+ },
751
+ "cost": {
752
+ "input": 0.05,
753
+ "output": 0.1
754
+ }
755
+ },
724
756
  "google/gemma-4-31B-it": {
725
757
  "id": "google/gemma-4-31B-it",
726
758
  "name": "Gemma 4 31B IT",
@@ -795,6 +827,38 @@
795
827
  "output": 0.1
796
828
  }
797
829
  },
830
+ "google/gemma-3-27b-it": {
831
+ "id": "google/gemma-3-27b-it",
832
+ "name": "Gemma 3 27B IT",
833
+ "description": "Largest open Gemma 3 instruction model for multilingual text generation and visual understanding",
834
+ "family": "gemma",
835
+ "attachment": true,
836
+ "reasoning": false,
837
+ "tool_call": true,
838
+ "structured_output": true,
839
+ "temperature": true,
840
+ "knowledge": "2024-08",
841
+ "release_date": "2025-03-12",
842
+ "last_updated": "2025-03-12",
843
+ "modalities": {
844
+ "input": [
845
+ "text",
846
+ "image"
847
+ ],
848
+ "output": [
849
+ "text"
850
+ ]
851
+ },
852
+ "open_weights": true,
853
+ "limit": {
854
+ "context": 131072,
855
+ "output": 131072
856
+ },
857
+ "cost": {
858
+ "input": 0.08,
859
+ "output": 0.16
860
+ }
861
+ },
798
862
  "google/gemma-4-26B-A4B-it": {
799
863
  "id": "google/gemma-4-26B-A4B-it",
800
864
  "name": "Gemma 4 26B A4B IT",
@@ -831,6 +895,38 @@
831
895
  "output": 0.34
832
896
  }
833
897
  },
898
+ "google/gemma-3-12b-it": {
899
+ "id": "google/gemma-3-12b-it",
900
+ "name": "Gemma 3 12B IT",
901
+ "description": "Open multimodal Gemma instruction model for multilingual text generation and image understanding",
902
+ "family": "gemma",
903
+ "attachment": true,
904
+ "reasoning": false,
905
+ "tool_call": true,
906
+ "structured_output": true,
907
+ "temperature": true,
908
+ "knowledge": "2024-08",
909
+ "release_date": "2025-03-12",
910
+ "last_updated": "2025-03-12",
911
+ "modalities": {
912
+ "input": [
913
+ "text",
914
+ "image"
915
+ ],
916
+ "output": [
917
+ "text"
918
+ ]
919
+ },
920
+ "open_weights": true,
921
+ "limit": {
922
+ "context": 131072,
923
+ "output": 131072
924
+ },
925
+ "cost": {
926
+ "input": 0.05,
927
+ "output": 0.15
928
+ }
929
+ },
834
930
  "zai-org/GLM-5.1": {
835
931
  "id": "zai-org/GLM-5.1",
836
932
  "name": "GLM-5.1",
@@ -909,7 +1005,7 @@
909
1005
  "cost": {
910
1006
  "input": 1.2,
911
1007
  "output": 4,
912
- "cache_read": 0.12
1008
+ "cache_read": 0.2
913
1009
  }
914
1010
  },
915
1011
  "zai-org/GLM-5.2": {
@@ -990,6 +1086,7 @@
990
1086
  "context": 202752,
991
1087
  "output": 16384
992
1088
  },
1089
+ "status": "deprecated",
993
1090
  "cost": {
994
1091
  "input": 0.06,
995
1092
  "output": 0.4,
@@ -1070,6 +1167,7 @@
1070
1167
  "context": 202752,
1071
1168
  "output": 16384
1072
1169
  },
1170
+ "status": "deprecated",
1073
1171
  "cost": {
1074
1172
  "input": 0.6,
1075
1173
  "output": 2.08,
@@ -1120,7 +1218,7 @@
1120
1218
  "id": "zai-org/GLM-5.3-Flash",
1121
1219
  "name": "GLM-5.3-Flash",
1122
1220
  "description": "Native multimodal GLM model for efficient coding and long-horizon agent tasks",
1123
- "family": "glm",
1221
+ "family": "glm-flash",
1124
1222
  "attachment": true,
1125
1223
  "reasoning": true,
1126
1224
  "reasoning_options": [
@@ -1306,7 +1404,8 @@
1306
1404
  "modalities": {
1307
1405
  "input": [
1308
1406
  "text",
1309
- "image"
1407
+ "image",
1408
+ "video"
1310
1409
  ],
1311
1410
  "output": [
1312
1411
  "text"
@@ -1318,9 +1417,9 @@
1318
1417
  "output": 32768
1319
1418
  },
1320
1419
  "cost": {
1321
- "input": 0.4,
1322
- "output": 3,
1323
- "cache_read": 0.04
1420
+ "input": 0.2,
1421
+ "output": 2.5,
1422
+ "cache_read": 0.05
1324
1423
  }
1325
1424
  },
1326
1425
  "Qwen/Qwen3.5-27B": {
@@ -1592,6 +1691,38 @@
1592
1691
  "output": 0.55
1593
1692
  }
1594
1693
  },
1694
+ "Qwen/Qwen3.8-Flash": {
1695
+ "id": "Qwen/Qwen3.8-Flash",
1696
+ "name": "Qwen3.8 Flash",
1697
+ "description": "Qwen vision-language model for visual reasoning, documents, and agent tasks",
1698
+ "family": "qwen",
1699
+ "attachment": true,
1700
+ "reasoning": false,
1701
+ "tool_call": true,
1702
+ "structured_output": true,
1703
+ "release_date": "2026-08-26",
1704
+ "last_updated": "2026-08-26",
1705
+ "modalities": {
1706
+ "input": [
1707
+ "text",
1708
+ "image",
1709
+ "video"
1710
+ ],
1711
+ "output": [
1712
+ "text"
1713
+ ]
1714
+ },
1715
+ "open_weights": false,
1716
+ "limit": {
1717
+ "context": 1000000,
1718
+ "output": 131072
1719
+ },
1720
+ "cost": {
1721
+ "input": 0.113,
1722
+ "output": 0.382,
1723
+ "cache_read": 0.0141
1724
+ }
1725
+ },
1595
1726
  "Qwen/Qwen3-VL-235B-A22B-Instruct": {
1596
1727
  "id": "Qwen/Qwen3-VL-235B-A22B-Instruct",
1597
1728
  "name": "Qwen3 VL 235B A22B Instruct",
@@ -1972,6 +2103,7 @@
1972
2103
  "context": 196608,
1973
2104
  "output": 131072
1974
2105
  },
2106
+ "status": "deprecated",
1975
2107
  "cost": {
1976
2108
  "input": 0.25,
1977
2109
  "output": 1,
@@ -2344,9 +2476,9 @@
2344
2476
  "output": 128000
2345
2477
  },
2346
2478
  "cost": {
2347
- "input": 0.14,
2348
- "output": 0.58,
2349
- "cache_read": 0.035
2479
+ "input": 0.13,
2480
+ "output": 0.53,
2481
+ "cache_read": 0.033
2350
2482
  }
2351
2483
  },
2352
2484
  "XiaomiMiMo/MiMo-V2.5-Pro": {
@@ -2428,9 +2560,9 @@
2428
2560
  "output": 16384
2429
2561
  },
2430
2562
  "cost": {
2431
- "input": 0.4,
2432
- "output": 2,
2433
- "cache_read": 0.08
2563
+ "input": 0.14,
2564
+ "output": 0.28,
2565
+ "cache_read": 0.0028
2434
2566
  }
2435
2567
  }
2436
2568
  }
data/data/deepseek.json CHANGED
@@ -51,6 +51,7 @@
51
51
  "context": 1000000,
52
52
  "output": 384000
53
53
  },
54
+ "status": "deprecated",
54
55
  "cost": {
55
56
  "input": 0.15,
56
57
  "output": 0.6,
@@ -101,6 +102,7 @@
101
102
  "context": 1000000,
102
103
  "output": 384000
103
104
  },
105
+ "status": "deprecated",
104
106
  "cost": {
105
107
  "input": 0.15,
106
108
  "output": 0.6,
@@ -122,6 +124,7 @@
122
124
  {
123
125
  "type": "effort",
124
126
  "values": [
127
+ "low",
125
128
  "high",
126
129
  "max"
127
130
  ]
data/data/mistral.json CHANGED
@@ -285,6 +285,47 @@
285
285
  "output": 5
286
286
  }
287
287
  },
288
+ "zai-glm-5-3": {
289
+ "id": "zai-glm-5-3",
290
+ "name": "GLM-5.3",
291
+ "description": "Flagship GLM model for long-horizon coding, agents, and complex project delivery",
292
+ "family": "glm",
293
+ "attachment": false,
294
+ "reasoning": true,
295
+ "reasoning_options": [
296
+ {
297
+ "type": "effort",
298
+ "values": [
299
+ "low",
300
+ "high",
301
+ "max"
302
+ ]
303
+ }
304
+ ],
305
+ "tool_call": true,
306
+ "structured_output": true,
307
+ "temperature": true,
308
+ "release_date": "2026-08-14",
309
+ "last_updated": "2026-08-14",
310
+ "modalities": {
311
+ "input": [
312
+ "text"
313
+ ],
314
+ "output": [
315
+ "text"
316
+ ]
317
+ },
318
+ "open_weights": true,
319
+ "limit": {
320
+ "context": 1000000,
321
+ "output": 131072
322
+ },
323
+ "cost": {
324
+ "input": 1.4,
325
+ "output": 4.4,
326
+ "cache_read": 0.14
327
+ }
328
+ },
288
329
  "open-mixtral-8x22b": {
289
330
  "id": "open-mixtral-8x22b",
290
331
  "name": "Mixtral 8x22B",
@@ -862,6 +903,7 @@
862
903
  {
863
904
  "type": "effort",
864
905
  "values": [
906
+ "none",
865
907
  "high",
866
908
  "max"
867
909
  ]
data/data/openai.json CHANGED
@@ -160,6 +160,99 @@
160
160
  "output": 120
161
161
  }
162
162
  },
163
+ "gpt-6-sol": {
164
+ "id": "gpt-6-sol",
165
+ "name": "GPT-6 Sol",
166
+ "description": "OpenAI model for complex coding and agentic workflows",
167
+ "family": "gpt-sol",
168
+ "attachment": true,
169
+ "reasoning": true,
170
+ "reasoning_options": [
171
+ {
172
+ "type": "effort",
173
+ "values": [
174
+ "none",
175
+ "low",
176
+ "medium",
177
+ "high",
178
+ "xhigh",
179
+ "max"
180
+ ]
181
+ }
182
+ ],
183
+ "tool_call": true,
184
+ "structured_output": true,
185
+ "temperature": false,
186
+ "knowledge": "2026-04-20",
187
+ "release_date": "2026-09-22",
188
+ "last_updated": "2026-09-22",
189
+ "modalities": {
190
+ "input": [
191
+ "text",
192
+ "image",
193
+ "pdf"
194
+ ],
195
+ "output": [
196
+ "text"
197
+ ]
198
+ },
199
+ "open_weights": false,
200
+ "limit": {
201
+ "context": 1050000,
202
+ "input": 922000,
203
+ "output": 128000
204
+ },
205
+ "experimental": {
206
+ "modes": {
207
+ "fast": {
208
+ "cost": {
209
+ "input": 4,
210
+ "output": 20,
211
+ "cache_read": 0.4,
212
+ "cache_write": 5
213
+ },
214
+ "provider": {
215
+ "body": {
216
+ "service_tier": "priority"
217
+ }
218
+ }
219
+ },
220
+ "pro": {
221
+ "provider": {
222
+ "body": {
223
+ "reasoning": {
224
+ "mode": "pro"
225
+ }
226
+ }
227
+ }
228
+ }
229
+ }
230
+ },
231
+ "cost": {
232
+ "input": 2,
233
+ "output": 10,
234
+ "cache_read": 0.2,
235
+ "cache_write": 2.5,
236
+ "tiers": [
237
+ {
238
+ "input": 4,
239
+ "output": 15,
240
+ "cache_read": 0.4,
241
+ "cache_write": 5,
242
+ "tier": {
243
+ "type": "context",
244
+ "size": 272000
245
+ }
246
+ }
247
+ ],
248
+ "context_over_200k": {
249
+ "input": 4,
250
+ "output": 15,
251
+ "cache_read": 0.4,
252
+ "cache_write": 5
253
+ }
254
+ }
255
+ },
163
256
  "chatgpt-image-latest": {
164
257
  "id": "chatgpt-image-latest",
165
258
  "name": "chatgpt-image-latest",
@@ -766,6 +859,99 @@
766
859
  "cache_read": 1.25
767
860
  }
768
861
  },
862
+ "gpt-6-luna": {
863
+ "id": "gpt-6-luna",
864
+ "name": "GPT-6 Luna",
865
+ "description": "OpenAI's most efficient model for focused, high-volume tasks",
866
+ "family": "gpt-luna",
867
+ "attachment": true,
868
+ "reasoning": true,
869
+ "reasoning_options": [
870
+ {
871
+ "type": "effort",
872
+ "values": [
873
+ "none",
874
+ "low",
875
+ "medium",
876
+ "high",
877
+ "xhigh",
878
+ "max"
879
+ ]
880
+ }
881
+ ],
882
+ "tool_call": true,
883
+ "structured_output": true,
884
+ "temperature": false,
885
+ "knowledge": "2026-05-18",
886
+ "release_date": "2026-09-22",
887
+ "last_updated": "2026-09-22",
888
+ "modalities": {
889
+ "input": [
890
+ "text",
891
+ "image",
892
+ "pdf"
893
+ ],
894
+ "output": [
895
+ "text"
896
+ ]
897
+ },
898
+ "open_weights": false,
899
+ "limit": {
900
+ "context": 1050000,
901
+ "input": 922000,
902
+ "output": 128000
903
+ },
904
+ "experimental": {
905
+ "modes": {
906
+ "fast": {
907
+ "cost": {
908
+ "input": 0.2,
909
+ "output": 1,
910
+ "cache_read": 0.02,
911
+ "cache_write": 0.25
912
+ },
913
+ "provider": {
914
+ "body": {
915
+ "service_tier": "priority"
916
+ }
917
+ }
918
+ },
919
+ "pro": {
920
+ "provider": {
921
+ "body": {
922
+ "reasoning": {
923
+ "mode": "pro"
924
+ }
925
+ }
926
+ }
927
+ }
928
+ }
929
+ },
930
+ "cost": {
931
+ "input": 0.1,
932
+ "output": 0.5,
933
+ "cache_read": 0.01,
934
+ "cache_write": 0.125,
935
+ "tiers": [
936
+ {
937
+ "input": 0.2,
938
+ "output": 0.75,
939
+ "cache_read": 0.02,
940
+ "cache_write": 0.25,
941
+ "tier": {
942
+ "type": "context",
943
+ "size": 272000
944
+ }
945
+ }
946
+ ],
947
+ "context_over_200k": {
948
+ "input": 0.2,
949
+ "output": 0.75,
950
+ "cache_read": 0.02,
951
+ "cache_write": 0.25
952
+ }
953
+ }
954
+ },
769
955
  "gpt-5.6-luna": {
770
956
  "id": "gpt-5.6-luna",
771
957
  "name": "GPT-5.6 Luna",