@linxiraos/pi-catalog 1.4.6 → 1.4.9

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. package/CHANGELOG.md +6 -2
  2. package/README.md +3 -3
  3. package/dist/types/build.d.ts +12 -4
  4. package/dist/types/compat/auth-ids.d.ts +2 -2
  5. package/dist/types/compat/catalog-policy.d.ts +6 -4
  6. package/dist/types/compat/factory-droid.d.ts +7 -1
  7. package/dist/types/compat/provider-ids.d.ts +1 -1
  8. package/dist/types/compat/resolve.d.ts +1 -1
  9. package/dist/types/compat/revision.d.ts +2 -0
  10. package/dist/types/compat/server-side-fallback.d.ts +6 -4
  11. package/dist/types/compat/types.d.ts +6 -2
  12. package/dist/types/discovery/codex.d.ts +4 -5
  13. package/dist/types/discovery/cursor-proto.d.ts +8 -0
  14. package/dist/types/discovery/openai-compatible.d.ts +6 -0
  15. package/dist/types/model-cache.d.ts +8 -0
  16. package/dist/types/model-catalog-format.d.ts +37 -0
  17. package/dist/types/model-manager.d.ts +11 -1
  18. package/dist/types/model-thinking.d.ts +4 -0
  19. package/dist/types/models.d.ts +22 -3
  20. package/dist/types/pricing.d.ts +66 -6
  21. package/dist/types/provider-models/openai-compat.d.ts +18 -13
  22. package/dist/types/types.d.ts +14 -19
  23. package/dist/types/utils.d.ts +8 -0
  24. package/dist/types/wire/codex.d.ts +1 -1
  25. package/package.json +6 -10
  26. package/src/build.d.ts +12 -4
  27. package/src/build.ts +50 -3
  28. package/src/compat/auth-ids.ts +5 -0
  29. package/src/compat/axes.ts +27 -9
  30. package/src/compat/catalog-policy.d.ts +9 -0
  31. package/src/compat/catalog-policy.ts +14 -10
  32. package/src/compat/collapse.ts +1 -1
  33. package/src/compat/factory-droid.d.ts +7 -1
  34. package/src/compat/factory-droid.ts +9 -1
  35. package/src/compat/provider-ids.d.ts +1 -1
  36. package/src/compat/provider-ids.ts +2 -0
  37. package/src/compat/resolve.d.ts +1 -1
  38. package/src/compat/resolve.ts +4 -5
  39. package/src/compat/revision.d.ts +2 -0
  40. package/src/compat/revision.ts +10 -0
  41. package/src/compat/rules/README.md +14 -12
  42. package/src/compat/rules/auth/_order.kdl +1 -1
  43. package/src/compat/rules/auth/mistral.kdl +4 -0
  44. package/src/compat/rules/auth/modular.kdl +9 -0
  45. package/src/compat/rules/auth/scaleway.kdl +17 -0
  46. package/src/compat/rules/auth/typesafe.kdl +2 -1
  47. package/src/compat/rules/classes/anthropic.kdl +5 -0
  48. package/src/compat/rules/classes/deepseek.kdl +3 -2
  49. package/src/compat/rules/classes/gemini.kdl +1 -1
  50. package/src/compat/rules/classes/openai.kdl +2 -2
  51. package/src/compat/rules/classes/xai.kdl +1 -1
  52. package/src/compat/rules/providers/anthropic.kdl +8 -4
  53. package/src/compat/rules/providers/cursor.kdl +10 -0
  54. package/src/compat/rules/providers/deepseek.kdl +2 -2
  55. package/src/compat/rules/providers/factory-droid.kdl +243 -13
  56. package/src/compat/rules/providers/google-vertex.kdl +3 -0
  57. package/src/compat/rules/providers/kimi-code.kdl +6 -0
  58. package/src/compat/rules/providers/mistral.kdl +8 -0
  59. package/src/compat/rules/providers/modular.kdl +35 -0
  60. package/src/compat/rules/providers/openai-codex.kdl +1 -1
  61. package/src/compat/rules/providers/sambanova.kdl +2 -0
  62. package/src/compat/rules/providers/scaleway.kdl +179 -0
  63. package/src/compat/rules/runtime/behavior.kdl +22 -6
  64. package/src/compat/rules.json +1 -1
  65. package/src/compat/server-side-fallback.d.ts +9 -0
  66. package/src/compat/server-side-fallback.ts +7 -5
  67. package/src/compat/types.d.ts +6 -2
  68. package/src/compat/types.ts +6 -2
  69. package/src/discovery/codex.d.ts +4 -5
  70. package/src/discovery/codex.ts +7 -15
  71. package/src/discovery/cursor-proto.d.ts +8 -0
  72. package/src/discovery/cursor-proto.ts +12 -1
  73. package/src/discovery/factory-droid.ts +35 -11
  74. package/src/discovery/openai-compatible.d.ts +6 -0
  75. package/src/discovery/openai-compatible.ts +1 -1
  76. package/src/model-cache.d.ts +8 -0
  77. package/src/model-cache.ts +83 -61
  78. package/src/model-catalog-format.d.ts +37 -0
  79. package/src/model-catalog-format.ts +296 -0
  80. package/src/model-manager.d.ts +11 -1
  81. package/src/model-manager.ts +85 -143
  82. package/src/model-thinking.d.ts +4 -0
  83. package/src/model-thinking.ts +10 -0
  84. package/src/model-tokenizer.ts +13 -12
  85. package/src/models.d.ts +22 -3
  86. package/src/models.json +1 -1
  87. package/src/models.ts +48 -11
  88. package/src/pricing.d.ts +66 -6
  89. package/src/pricing.ts +59 -94
  90. package/src/provider-models/cache-provider-id.ts +23 -6
  91. package/src/provider-models/catalog-snapshot.json.zst +0 -0
  92. package/src/provider-models/catalog-snapshot.json.zst.d.ts +3 -0
  93. package/src/provider-models/descriptors.ts +2 -0
  94. package/src/provider-models/openai-compat.d.ts +18 -13
  95. package/src/provider-models/openai-compat.ts +203 -94
  96. package/src/provider-models/special.ts +5 -6
  97. package/src/types.d.ts +14 -19
  98. package/src/types.ts +11 -18
  99. package/src/utils.d.ts +8 -0
  100. package/src/utils.ts +18 -0
  101. package/src/wire/codex.d.ts +1 -1
  102. package/src/wire/codex.ts +1 -1
  103. package/src/wire/factory-droid.ts +4 -3
  104. package/src/models.json.d.ts +0 -9
@@ -141,24 +141,54 @@ provider "factory-droid" {
141
141
  cost input=0 output=0 cache-read=0 cache-write=0
142
142
  limits context=922000 max-tokens=128000
143
143
  }
144
+ model "gpt-6-astra-ultrafast" name="GPT-6 Astra Ultra Fast" {
145
+ reasoning #true
146
+ input "text" "image"
147
+ cost input=0 output=0 cache-read=0 cache-write=0
148
+ limits context=922000 max-tokens=128000
149
+ }
144
150
  model "gpt-6.1-sol" name="GPT-6.1 Sol" {
145
151
  reasoning #true
146
152
  input "text" "image"
147
153
  cost input=0 output=0 cache-read=0 cache-write=0
148
154
  limits context=922000 max-tokens=128000
149
155
  }
156
+ model "gpt-6.1-sol-ultrafast" name="GPT-6.1 Sol Ultra Fast" {
157
+ reasoning #true
158
+ input "text" "image"
159
+ cost input=0 output=0 cache-read=0 cache-write=0
160
+ limits context=922000 max-tokens=128000
161
+ }
162
+ model "gpt-6.1-terra" name="GPT-6.1 Terra" {
163
+ reasoning #true
164
+ input "text" "image"
165
+ cost input=0 output=0 cache-read=0 cache-write=0
166
+ limits context=922000 max-tokens=128000
167
+ }
150
168
  model "gpt-6-sol" name="GPT-6 Sol" {
151
169
  reasoning #true
152
170
  input "text" "image"
153
171
  cost input=0 output=0 cache-read=0 cache-write=0
154
172
  limits context=922000 max-tokens=128000
155
173
  }
174
+ model "gpt-6-sol-fast" name="GPT-6 Sol Fast Mode" {
175
+ reasoning #true
176
+ input "text" "image"
177
+ cost input=0 output=0 cache-read=0 cache-write=0
178
+ limits context=922000 max-tokens=128000
179
+ }
156
180
  model "gpt-6-luna" name="GPT-6 Luna" {
157
181
  reasoning #true
158
182
  input "text" "image"
159
183
  cost input=0 output=0 cache-read=0 cache-write=0
160
184
  limits context=922000 max-tokens=128000
161
185
  }
186
+ model "gpt-6-luna-fast" name="GPT-6 Luna Fast Mode" {
187
+ reasoning #true
188
+ input "text" "image"
189
+ cost input=0 output=0 cache-read=0 cache-write=0
190
+ limits context=922000 max-tokens=128000
191
+ }
162
192
  model "gpt-5.6-sol" name="GPT-5.6 Sol" {
163
193
  reasoning #true
164
194
  input "text" "image"
@@ -339,12 +369,42 @@ provider "factory-droid" {
339
369
  cost input=0 output=0 cache-read=0 cache-write=0
340
370
  limits context=131072 max-tokens=131072
341
371
  }
372
+ model "nemotron-3.5-super" name="Nemotron 3.5 Super" {
373
+ reasoning #true
374
+ input "text" "image"
375
+ cost input=0 output=0 cache-read=0 cache-write=0
376
+ limits context=934464 max-tokens=65536
377
+ }
342
378
  model "nemotron-3-ultra" name="Nemotron 3 Ultra" {
343
379
  reasoning #true
344
380
  input "text"
345
381
  cost input=0 output=0 cache-read=0 cache-write=0
346
382
  limits context=136464 max-tokens=65536
347
383
  }
384
+ model "muse-spark-open" name="Muse Spark Open" {
385
+ reasoning #true
386
+ input "text" "image"
387
+ cost input=0 output=0 cache-read=0 cache-write=0
388
+ limits context=908928 max-tokens=131072
389
+ }
390
+ model "muse-spark-1.2" name="Muse Spark 1.2" {
391
+ reasoning #true
392
+ input "text"
393
+ cost input=0 output=0 cache-read=0 cache-write=0
394
+ limits context=908928 max-tokens=131072
395
+ }
396
+ model "beam-1.0" name="Beam 1.0" {
397
+ reasoning #true
398
+ input "text"
399
+ cost input=0 output=0 cache-read=0 cache-write=0
400
+ limits context=1015808 max-tokens=32768
401
+ }
402
+ model "deepseek-v4.1-flash-1017" name="DeepSeek V4.1 Flash 1017" {
403
+ reasoning #true
404
+ input "text" "image"
405
+ cost input=0 output=0 cache-read=0 cache-write=0
406
+ limits context=908928 max-tokens=131072
407
+ }
348
408
  model "deepseek-v4.1-flash" name="DeepSeek V4.1 Flash" {
349
409
  reasoning #true
350
410
  input "text" "image"
@@ -440,7 +500,10 @@ provider "factory-droid" {
440
500
  // Per-model registry entries: live-routing family, upstream rotation (first
441
501
  // entry is the default `x-api-provider`), region overrides, Standard Credits
442
502
  // rates, list-price source, org-policy aliases, the effort ladder (an off/none
443
- // rung means `thinking-requires-effort #false`), and account gates.
503
+ // rung means `thinking-requires-effort #false`), and account gates. A credit
504
+ // promotion is two dated `effective-rates` cards copied from the droid
505
+ // release's model registry: the discounted `input` from its start and the
506
+ // base again from its end.
444
507
  models "claude-fable-5.1" {
445
508
  routing-family "anthropic"
446
509
  upstream-rotation "anthropic" "vertex_anthropic" "bedrock_anthropic"
@@ -478,7 +541,7 @@ provider "factory-droid" {
478
541
  thinking-efforts "low" "medium" "high" "xhigh" "max"
479
542
  thinking-requires-effort #true
480
543
  thinking-default-level "medium"
481
- entitlement { feature-flag "claude_opus_5_5_fast"; base-variant "claude-opus-5-5" }
544
+ entitlement { base-variant "claude-opus-5-5" }
482
545
  }
483
546
  models "claude-opus-5" {
484
547
  routing-family "anthropic"
@@ -615,6 +678,20 @@ provider "factory-droid" {
615
678
  thinking-requires-effort #true
616
679
  thinking-default-level "medium"
617
680
  }
681
+ models "gpt-6-astra-ultrafast" {
682
+ routing-family "openai"
683
+ upstream-rotation "openai"
684
+ region-upstreams-eu
685
+ credit-rates { input 24; output 5 }
686
+ thinking-efforts "low" "medium" "high" "xhigh" "max"
687
+ thinking-requires-effort #true
688
+ thinking-default-level "medium"
689
+ entitlement {
690
+ feature-flag "gpt_6_astra_ultrafast"
691
+ base-variant "gpt-6-astra"
692
+ requires-explicit-opt-in #true
693
+ }
694
+ }
618
695
  models "gpt-6.1-sol" {
619
696
  routing-family "openai"
620
697
  upstream-rotation "openai" "bedrock_openai" "databricks"
@@ -623,7 +700,28 @@ provider "factory-droid" {
623
700
  thinking-efforts "low" "medium" "high" "xhigh" "max"
624
701
  thinking-requires-effort #true
625
702
  thinking-default-level "medium"
626
- entitlement { feature-flag "gpt_6_1_sol" }
703
+ }
704
+ models "gpt-6.1-sol-ultrafast" {
705
+ routing-family "openai"
706
+ upstream-rotation "openai"
707
+ credit-rates { input 4.8; output 5; cache-read 0.05 }
708
+ thinking-efforts "low" "medium" "high" "xhigh" "max"
709
+ thinking-requires-effort #true
710
+ thinking-default-level "medium"
711
+ entitlement {
712
+ base-variant "gpt-6.1-sol"
713
+ requires-explicit-opt-in #true
714
+ }
715
+ }
716
+ models "gpt-6.1-terra" {
717
+ routing-family "openai"
718
+ upstream-rotation "openai"
719
+ credit-rates { input 0.4; output 5; cache-read 0.05 }
720
+ list-price-from "openai"
721
+ thinking-efforts "low" "medium" "high" "xhigh" "max"
722
+ thinking-requires-effort #false
723
+ thinking-default-level "medium"
724
+ entitlement { feature-flag "gpt_6_1_terra" }
627
725
  }
628
726
  models "gpt-6-sol" {
629
727
  routing-family "openai"
@@ -635,6 +733,15 @@ provider "factory-droid" {
635
733
  thinking-requires-effort #false
636
734
  thinking-default-level "medium"
637
735
  }
736
+ models "gpt-6-sol-fast" {
737
+ routing-family "openai"
738
+ upstream-rotation "openai"
739
+ credit-rates { input 1.6; output 5 }
740
+ thinking-efforts "low" "medium" "high" "xhigh" "max"
741
+ thinking-requires-effort #false
742
+ thinking-default-level "medium"
743
+ entitlement { base-variant "gpt-6-sol" }
744
+ }
638
745
  models "gpt-6-luna" {
639
746
  routing-family "openai"
640
747
  upstream-rotation "openai" "bedrock_openai" "azure_openai"
@@ -645,11 +752,33 @@ provider "factory-droid" {
645
752
  thinking-requires-effort #false
646
753
  thinking-default-level "medium"
647
754
  }
755
+ models "gpt-6-luna-fast" {
756
+ routing-family "openai"
757
+ upstream-rotation "openai"
758
+ credit-rates { input 0.08; output 5 }
759
+ thinking-efforts "low" "medium" "high" "xhigh" "max"
760
+ thinking-requires-effort #false
761
+ thinking-default-level "medium"
762
+ entitlement { base-variant "gpt-6-luna" }
763
+ }
648
764
  models "gpt-5.6-sol" {
649
765
  routing-family "openai"
650
766
  upstream-rotation "openai" "bedrock_openai" "azure_openai" "snowflake" "databricks"
651
767
  region-upstreams-eu "openai"
652
- credit-rates { input 2; output 5 }
768
+ credit-rates {
769
+ input 2
770
+ output 5
771
+ effective-rates {
772
+ promotion {
773
+ effective-from "2026-08-22T00:00:00Z"
774
+ input 1.6
775
+ }
776
+ standard {
777
+ effective-from "2026-11-22T00:00:00Z"
778
+ input 2
779
+ }
780
+ }
781
+ }
653
782
  list-price-from "openai"
654
783
  thinking-efforts "low" "medium" "high" "xhigh" "max"
655
784
  thinking-requires-effort #false
@@ -658,7 +787,20 @@ provider "factory-droid" {
658
787
  models "gpt-5.6-sol-fast" {
659
788
  routing-family "openai"
660
789
  upstream-rotation "openai" "azure_openai"
661
- credit-rates { input 4; output 5 }
790
+ credit-rates {
791
+ input 4
792
+ output 5
793
+ effective-rates {
794
+ promotion {
795
+ effective-from "2026-08-22T00:00:00Z"
796
+ input 3.2
797
+ }
798
+ standard {
799
+ effective-from "2026-11-22T00:00:00Z"
800
+ input 4
801
+ }
802
+ }
803
+ }
662
804
  policy-aliases "gpt-5.6-fast"
663
805
  thinking-efforts "low" "medium" "high" "xhigh" "max"
664
806
  thinking-requires-effort #false
@@ -687,7 +829,7 @@ provider "factory-droid" {
687
829
  }
688
830
  models "gpt-5.5" {
689
831
  routing-family "openai"
690
- upstream-rotation "openai" "bedrock_openai" "azure_openai" "snowflake" "databricks"
832
+ upstream-rotation "openai" "bedrock_openai" "azure_openai" "databricks"
691
833
  region-upstreams-eu "openai"
692
834
  credit-rates { input 2; output 6 }
693
835
  list-price-from "openai"
@@ -810,7 +952,20 @@ provider "factory-droid" {
810
952
  routing-family "google"
811
953
  upstream-rotation "google"
812
954
  region-upstreams-eu "google"
813
- credit-rates { input 0.6; output 5 }
955
+ credit-rates {
956
+ input 0.6
957
+ output 5
958
+ effective-rates {
959
+ promotion {
960
+ effective-from "2026-08-17T21:10:19Z"
961
+ input 0.3
962
+ }
963
+ standard {
964
+ effective-from "2027-01-01T08:00:00Z"
965
+ input 0.6
966
+ }
967
+ }
968
+ }
814
969
  list-price-from "google"
815
970
  thinking-efforts "low" "medium" "high"
816
971
  thinking-requires-effort #true
@@ -820,7 +975,20 @@ provider "factory-droid" {
820
975
  routing-family "google"
821
976
  upstream-rotation "google"
822
977
  region-upstreams-eu "google"
823
- credit-rates { input 0.6; output 5 }
978
+ credit-rates {
979
+ input 0.6
980
+ output 5
981
+ effective-rates {
982
+ promotion {
983
+ effective-from "2026-08-17T21:10:19Z"
984
+ input 0.3
985
+ }
986
+ standard {
987
+ effective-from "2027-01-01T08:00:00Z"
988
+ input 0.6
989
+ }
990
+ }
991
+ }
824
992
  list-price-from "google"
825
993
  thinking-efforts "low" "medium" "high"
826
994
  thinking-requires-effort #true
@@ -966,6 +1134,15 @@ provider "factory-droid" {
966
1134
  thinking-requires-effort #true
967
1135
  thinking-default-level "xhigh"
968
1136
  }
1137
+ models "nemotron-3.5-super" {
1138
+ routing-family "factory"
1139
+ upstream-rotation "baseten"
1140
+ credit-rates { input 0.08; output 2.5; cache-read 0.2 }
1141
+ thinking-efforts "high"
1142
+ thinking-requires-effort #false
1143
+ thinking-default-level "high"
1144
+ entitlement { feature-flag "nemotron_3_5_super" }
1145
+ }
969
1146
  models "nemotron-3-ultra" {
970
1147
  routing-family "factory"
971
1148
  upstream-rotation "baseten" "fireworks"
@@ -975,6 +1152,45 @@ provider "factory-droid" {
975
1152
  thinking-requires-effort #false
976
1153
  thinking-default-level "high"
977
1154
  }
1155
+ models "muse-spark-open" {
1156
+ routing-family "factory"
1157
+ upstream-rotation "baseten"
1158
+ credit-rates { input 0.5; output 3.4; cache-read 0.12 }
1159
+ thinking-efforts "low" "medium" "high" "xhigh" "max"
1160
+ thinking-requires-effort #true
1161
+ thinking-default-level "high"
1162
+ entitlement { feature-flag "muse_spark_open" }
1163
+ }
1164
+ // Factory serves no `minimal` rung, so this ladder outranks the Meta class ladder.
1165
+ models "muse-spark-1.2" priority=1 {
1166
+ routing-family "factory"
1167
+ upstream-rotation "fireworks"
1168
+ credit-rates { input 0.5; output 3.4; cache-read 0.12 }
1169
+ list-price-from "meta"
1170
+ thinking-efforts "low" "medium" "high" "xhigh"
1171
+ thinking-requires-effort #true
1172
+ thinking-default-level "high"
1173
+ }
1174
+ models "beam-1.0" {
1175
+ routing-family "factory"
1176
+ upstream-rotation "baseten"
1177
+ credit-rates { input 0.52; output 3.3077 }
1178
+ thinking-efforts "low" "medium" "high" "xhigh" "max"
1179
+ thinking-requires-effort #true
1180
+ thinking-default-level "high"
1181
+ entitlement { feature-flag "beam_1_0" }
1182
+ }
1183
+ models "deepseek-v4.1-flash-1017" {
1184
+ routing-family "factory"
1185
+ upstream-rotation "fireworks" "baseten" "databricks"
1186
+ credit-rates { input 0.12; output 4; cache-read 0.024 }
1187
+ // No bundled Fireworks row for the 1017 slug; Fireworks prices it like V4.1 Flash.
1188
+ list-price-from "fireworks" "deepseek-v4.1-flash"
1189
+ thinking-efforts "low" "high" "max"
1190
+ thinking-requires-effort #false
1191
+ thinking-default-level "high"
1192
+ entitlement { feature-flag "deepseek_v4_1_flash_1017" }
1193
+ }
978
1194
  models "deepseek-v4.1-flash" {
979
1195
  routing-family "factory"
980
1196
  upstream-rotation "fireworks" "baseten" "databricks"
@@ -1012,7 +1228,6 @@ provider "factory-droid" {
1012
1228
  thinking-efforts "high"
1013
1229
  thinking-requires-effort #true
1014
1230
  thinking-default-level "high"
1015
- entitlement { feature-flag "minimax_m3" }
1016
1231
  }
1017
1232
  models "grok-4.7" {
1018
1233
  routing-family "xai"
@@ -1092,8 +1307,8 @@ provider "factory-droid" {
1092
1307
  // Native declares every Responses tool `strict: false`; the field must reach the wire.
1093
1308
  supports-strict-mode #true
1094
1309
  }
1095
- models "glm-*" "kimi-*" "deepseek-*" "nemotron-*" "inkling" "mistral-medium-3.5" "mistral-large-4" "qwen3.8-max" \
1096
- "minimax-m3" {
1310
+ models "glm-*" "kimi-*" "deepseek-*" "nemotron-*" "muse-spark-open" "muse-spark-1.2" "beam-1.0" "inkling" \
1311
+ "mistral-medium-3.5" "mistral-large-4" "qwen3.8-max" "minimax-m3" {
1097
1312
  clamp-output-to-model-max #true
1098
1313
  completions-reasoning-mode "none"
1099
1314
  completions-reasoning-history "omit"
@@ -1119,6 +1334,10 @@ provider "factory-droid" {
1119
1334
  completions-reasoning-mode "opt-in"
1120
1335
  }
1121
1336
  }
1337
+ // Beam tolerates tool-call turns that arrive without reasoning.
1338
+ models "beam-1.0" {
1339
+ completions-allow-missing-thinking #true
1340
+ }
1122
1341
  models "deepseek-*" {
1123
1342
  on-upstream "baseten" priority=10 {
1124
1343
  completions-reasoning-mode "forced-on"
@@ -1131,6 +1350,13 @@ provider "factory-droid" {
1131
1350
  synthetic-reasoning-content-fallback " "
1132
1351
  }
1133
1352
  }
1353
+ // Droid 0.250.0 configures the 1017 checkpoint's Baseten route with
1354
+ // thinkingMode "reasoning-effort": thinking off is sent as `none`.
1355
+ models "deepseek-v4.1-flash-1017" {
1356
+ on-upstream "baseten" {
1357
+ completions-reasoning-mode "effort"
1358
+ }
1359
+ }
1134
1360
  // The replay codec follows the selected route, including EU GLM routes.
1135
1361
  on-upstream "mistral" {
1136
1362
  models "glm-*" "mistral-medium-3.5" "mistral-large-4" {
@@ -1217,7 +1443,7 @@ provider "factory-droid" {
1217
1443
  fast-mode #true
1218
1444
  }
1219
1445
  // Refusals retry server-side only on the first-party Anthropic upstream.
1220
- models "claude-fable-5" "claude-fable-5.1" {
1446
+ models "claude-fable-5" "claude-fable-5.1" "claude-opus-5-5" {
1221
1447
  on-upstream "anthropic" {
1222
1448
  server-side-fallback-models "claude-opus-5"
1223
1449
  }
@@ -1259,7 +1485,8 @@ provider "factory-droid" {
1259
1485
  }
1260
1486
  // GPT models whose ladder has an Off/None rung send the literal `none`
1261
1487
  // tier on every upstream.
1262
- models "gpt-5.2" "gpt-5.6-sol" "gpt-5.6-sol-fast" "gpt-5.6-terra" "gpt-5.6-luna" "gpt-6-sol" "gpt-6-luna" {
1488
+ models "gpt-5.2" "gpt-5.6-sol" "gpt-5.6-sol-fast" "gpt-5.6-terra" "gpt-5.6-luna" "gpt-6.1-terra" "gpt-6-sol" \
1489
+ "gpt-6-sol-fast" "gpt-6-luna" "gpt-6-luna-fast" {
1263
1490
  reasoning-disable-mode "none-effort"
1264
1491
  }
1265
1492
  models "gpt-*" {
@@ -1282,6 +1509,9 @@ provider "factory-droid" {
1282
1509
  models "gpt-*-fast" {
1283
1510
  responses-service-tier "priority"
1284
1511
  }
1512
+ models "gpt-*-ultrafast" {
1513
+ responses-service-tier "ultrafast"
1514
+ }
1285
1515
  models "grok-*" {
1286
1516
  supports-reasoning-summary #false
1287
1517
  }
@@ -4,6 +4,9 @@ provider "google-vertex" {
4
4
  default-model "gemini-3.1-pro-preview"
5
5
  allow-unauthenticated #true
6
6
  image-count-budget 200
7
+ // Vertex publishes Anthropic ids pinned to a revision (`claude-opus-4-8@default`);
8
+ // `google-vertex/claude-opus-4-8` names that model exactly.
9
+ selector-revision-suffix "@default"
7
10
 
8
11
  // Replaces google-shared.ts stripping functionCall/functionResponse ids for Vertex.
9
12
  supports-function-part-id #false
@@ -2,6 +2,7 @@
2
2
 
3
3
  provider "kimi-code" {
4
4
  default-model "kimi-for-coding"
5
+ dynamic-models-authoritative #true
5
6
  discovery label="Kimi Code" {
6
7
  env "KIMI_API_KEY"
7
8
  }
@@ -39,6 +40,11 @@ provider "kimi-code" {
39
40
  thinking-efforts "minimal" "low" "medium" "high"
40
41
  thinking-format "zai"
41
42
  }
43
+ // K2.7 Code Highspeed always thinks (`supports_thinking_type: "only"`) but
44
+ // exposes no adjustable effort ladder (`think_efforts` is omitted).
45
+ models "kimi-for-coding-highspeed" {
46
+ thinking-efforts
47
+ }
42
48
 
43
49
  // /coding/v1/models carries no pricing. `cost-fallback` seeds the public
44
50
  // Moonshot list price only when discovery reports $0, so a future priced
@@ -29,6 +29,14 @@ provider "mistral" {
29
29
  // without it, requests only hit replica-local implicit cache. Same shape as
30
30
  // the kimi-code rule (#6049).
31
31
  supports-prompt-cache-key #true
32
+ // The gateway accepts assistant reasoning as text, not DeepSeek's native
33
+ // reasoning_content field, even when it serves DeepSeek-family models.
34
+ requires-reasoning-content-for-tool-calls #false
35
+ requires-reasoning-content-for-all-assistant-turns #false
36
+ // Reasoning streams and replays as typed `thinking` content parts
37
+ // (https://docs.mistral.ai/studio/conversations/reasoning) for every
38
+ // reasoning model the host serves, including GLM and DeepSeek (#17375).
39
+ mistral-reasoning-content-parts #true
32
40
  class "unknown" {
33
41
  thinking-efforts "minimal" "low" "medium" "high" "xhigh"
34
42
  }
@@ -0,0 +1,35 @@
1
+ // Provider-wire compat for "modular". Modular Cloud serves open models on an
2
+ // OpenAI-compatible API. The provider is catalog-only: models, prices, and
3
+ // limits come from the models.dev `modular` entry. Reasoning controls differ
4
+ // per model and were measured against the API.
5
+
6
+ provider "modular" {
7
+ default-model "google/gemma-4-31b-it"
8
+ env "MODULAR_API_KEY"
9
+
10
+ thinking-mode "effort"
11
+ // The API takes OpenRouter's `reasoning` object for every model, so a model
12
+ // models.dev adds later works without its own block: on sends
13
+ // `reasoning.effort`, off sends `reasoning.enabled = false`.
14
+ thinking-format "openrouter"
15
+
16
+ // MiniMax M3.1 Flash takes a nested `reasoning.effort` and always thinks:
17
+ // `reasoning.enabled = false` still returns reasoning tokens.
18
+ models "MiniMaxAI/MiniMax-M3.1-Flash" {
19
+ thinking-requires-effort #true
20
+ thinking-efforts "low" "medium" "high" "xhigh" "max"
21
+ }
22
+ // MiniMax M3 and Gemma 4 31B switch reasoning with `reasoning.enabled` and
23
+ // ignore any effort, so the ladder is one on-rung beside off: off sends
24
+ // `enabled: false`, and on replaces the effort with `enabled: true`.
25
+ models "minimax/minimax-m3" "google/gemma-4-31b-it" {
26
+ thinking-efforts "high"
27
+ when-thinking {
28
+ extra-body {
29
+ reasoning {
30
+ enabled #true
31
+ }
32
+ }
33
+ }
34
+ }
35
+ }
@@ -16,7 +16,7 @@ provider "openai-codex" {
16
16
 
17
17
  // Policy budgets: the Codex backend documents no request-size or image
18
18
  // limit, and requests with several thousand images fail mid-stream at
19
- // random. Inline images keep Anthropic's 24 MB; the official endpoint's
19
+ // random. Inline images keep a 24 MB budget; the official endpoint's
20
20
  // image count stays well below those failures. A custom baseUrl keeps 200.
21
21
  inline-image-byte-budget 24000000
22
22
  image-count-budget 200
@@ -28,6 +28,8 @@ provider "sambanova" {
28
28
  // (https://docs.sambanova.ai/docs/api-reference/chat-completions/create-chat-based-completion).
29
29
  thinking-mode "effort"
30
30
  thinking-format "openai"
31
+ // Replay the streamed `reasoning` field so turn-N prompts hit the gateway's prefix cache.
32
+ replay-reasoning-content #true
31
33
  class "gpt-oss" {
32
34
  thinking-efforts "low" "medium" "high"
33
35
  thinking-upgrade-neutral #true