@dreb/ai 2.64.4 → 2.64.6

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.
@@ -13,11 +13,11 @@ export const MODELS = {
13
13
  cost: {
14
14
  input: 0.33,
15
15
  output: 2.75,
16
- cacheRead: 0,
17
- cacheWrite: 0,
16
+ cacheRead: 0.0825,
17
+ cacheWrite: 0.33,
18
18
  },
19
- contextWindow: 128000,
20
- maxTokens: 4096,
19
+ contextWindow: 1000000,
20
+ maxTokens: 65535,
21
21
  },
22
22
  "amazon.nova-lite-v1:0": {
23
23
  id: "amazon.nova-lite-v1:0",
@@ -31,10 +31,10 @@ export const MODELS = {
31
31
  input: 0.06,
32
32
  output: 0.24,
33
33
  cacheRead: 0.015,
34
- cacheWrite: 0,
34
+ cacheWrite: 0.06,
35
35
  },
36
36
  contextWindow: 300000,
37
- maxTokens: 8192,
37
+ maxTokens: 10000,
38
38
  },
39
39
  "amazon.nova-micro-v1:0": {
40
40
  id: "amazon.nova-micro-v1:0",
@@ -48,10 +48,10 @@ export const MODELS = {
48
48
  input: 0.035,
49
49
  output: 0.14,
50
50
  cacheRead: 0.00875,
51
- cacheWrite: 0,
51
+ cacheWrite: 0.035,
52
52
  },
53
53
  contextWindow: 128000,
54
- maxTokens: 8192,
54
+ maxTokens: 10000,
55
55
  },
56
56
  "amazon.nova-pro-v1:0": {
57
57
  id: "amazon.nova-pro-v1:0",
@@ -65,10 +65,10 @@ export const MODELS = {
65
65
  input: 0.8,
66
66
  output: 3.2,
67
67
  cacheRead: 0.2,
68
- cacheWrite: 0,
68
+ cacheWrite: 0.8,
69
69
  },
70
70
  contextWindow: 300000,
71
- maxTokens: 8192,
71
+ maxTokens: 10000,
72
72
  },
73
73
  "anthropic.claude-fable-5": {
74
74
  id: "anthropic.claude-fable-5",
@@ -274,6 +274,57 @@ export const MODELS = {
274
274
  contextWindow: 1000000,
275
275
  maxTokens: 128000,
276
276
  },
277
+ "apac.amazon.nova-lite-v1:0": {
278
+ id: "apac.amazon.nova-lite-v1:0",
279
+ name: "Nova Lite (APAC)",
280
+ api: "bedrock-converse-stream",
281
+ provider: "amazon-bedrock",
282
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
283
+ reasoning: false,
284
+ input: ["text", "image"],
285
+ cost: {
286
+ input: 0.063,
287
+ output: 0.252,
288
+ cacheRead: 0.01575,
289
+ cacheWrite: 0.063,
290
+ },
291
+ contextWindow: 300000,
292
+ maxTokens: 10000,
293
+ },
294
+ "apac.amazon.nova-micro-v1:0": {
295
+ id: "apac.amazon.nova-micro-v1:0",
296
+ name: "Nova Micro (APAC)",
297
+ api: "bedrock-converse-stream",
298
+ provider: "amazon-bedrock",
299
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
300
+ reasoning: false,
301
+ input: ["text"],
302
+ cost: {
303
+ input: 0.037,
304
+ output: 0.148,
305
+ cacheRead: 0.00925,
306
+ cacheWrite: 0.037,
307
+ },
308
+ contextWindow: 128000,
309
+ maxTokens: 10000,
310
+ },
311
+ "apac.amazon.nova-pro-v1:0": {
312
+ id: "apac.amazon.nova-pro-v1:0",
313
+ name: "Nova Pro (APAC)",
314
+ api: "bedrock-converse-stream",
315
+ provider: "amazon-bedrock",
316
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
317
+ reasoning: false,
318
+ input: ["text", "image"],
319
+ cost: {
320
+ input: 0.84,
321
+ output: 3.36,
322
+ cacheRead: 0.21,
323
+ cacheWrite: 0.84,
324
+ },
325
+ contextWindow: 300000,
326
+ maxTokens: 10000,
327
+ },
277
328
  "au.anthropic.claude-haiku-4-5-20251001-v1:0": {
278
329
  id: "au.anthropic.claude-haiku-4-5-20251001-v1:0",
279
330
  name: "Claude Haiku 4.5 (AU)",
@@ -308,6 +359,23 @@ export const MODELS = {
308
359
  contextWindow: 1000000,
309
360
  maxTokens: 128000,
310
361
  },
362
+ "au.anthropic.claude-opus-4-7": {
363
+ id: "au.anthropic.claude-opus-4-7",
364
+ name: "Claude Opus 4.7 (AU)",
365
+ api: "bedrock-converse-stream",
366
+ provider: "amazon-bedrock",
367
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
368
+ reasoning: true,
369
+ input: ["text", "image"],
370
+ cost: {
371
+ input: 5.5,
372
+ output: 27.5,
373
+ cacheRead: 0.55,
374
+ cacheWrite: 6.875,
375
+ },
376
+ contextWindow: 1000000,
377
+ maxTokens: 128000,
378
+ },
311
379
  "au.anthropic.claude-opus-4-8": {
312
380
  id: "au.anthropic.claude-opus-4-8",
313
381
  name: "Claude Opus 4.8 (AU)",
@@ -393,22 +461,22 @@ export const MODELS = {
393
461
  contextWindow: 1000000,
394
462
  maxTokens: 128000,
395
463
  },
396
- "deepseek.r1-v1:0": {
397
- id: "deepseek.r1-v1:0",
398
- name: "DeepSeek-R1",
464
+ "ca.amazon.nova-lite-v1:0": {
465
+ id: "ca.amazon.nova-lite-v1:0",
466
+ name: "Nova Lite (CA)",
399
467
  api: "bedrock-converse-stream",
400
468
  provider: "amazon-bedrock",
401
469
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
402
- reasoning: true,
403
- input: ["text"],
470
+ reasoning: false,
471
+ input: ["text", "image"],
404
472
  cost: {
405
- input: 1.35,
406
- output: 5.4,
407
- cacheRead: 0,
408
- cacheWrite: 0,
473
+ input: 0.064,
474
+ output: 0.256,
475
+ cacheRead: 0.016,
476
+ cacheWrite: 0.064,
409
477
  },
410
- contextWindow: 128000,
411
- maxTokens: 32768,
478
+ contextWindow: 300000,
479
+ maxTokens: 10000,
412
480
  },
413
481
  "deepseek.v3-v1:0": {
414
482
  id: "deepseek.v3-v1:0",
@@ -429,7 +497,7 @@ export const MODELS = {
429
497
  },
430
498
  "deepseek.v3.2": {
431
499
  id: "deepseek.v3.2",
432
- name: "DeepSeek-V3.2",
500
+ name: "DeepSeek V3.2",
433
501
  api: "bedrock-converse-stream",
434
502
  provider: "amazon-bedrock",
435
503
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
@@ -444,6 +512,74 @@ export const MODELS = {
444
512
  contextWindow: 163840,
445
513
  maxTokens: 81920,
446
514
  },
515
+ "eu.amazon.nova-2-lite-v1:0": {
516
+ id: "eu.amazon.nova-2-lite-v1:0",
517
+ name: "Nova 2 Lite (EU)",
518
+ api: "bedrock-converse-stream",
519
+ provider: "amazon-bedrock",
520
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
521
+ reasoning: true,
522
+ input: ["text", "image"],
523
+ cost: {
524
+ input: 0.374,
525
+ output: 3.157,
526
+ cacheRead: 0.0935,
527
+ cacheWrite: 0.374,
528
+ },
529
+ contextWindow: 1000000,
530
+ maxTokens: 65535,
531
+ },
532
+ "eu.amazon.nova-lite-v1:0": {
533
+ id: "eu.amazon.nova-lite-v1:0",
534
+ name: "Nova Lite (EU)",
535
+ api: "bedrock-converse-stream",
536
+ provider: "amazon-bedrock",
537
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
538
+ reasoning: false,
539
+ input: ["text", "image"],
540
+ cost: {
541
+ input: 0.069,
542
+ output: 0.276,
543
+ cacheRead: 0.01725,
544
+ cacheWrite: 0.069,
545
+ },
546
+ contextWindow: 300000,
547
+ maxTokens: 10000,
548
+ },
549
+ "eu.amazon.nova-micro-v1:0": {
550
+ id: "eu.amazon.nova-micro-v1:0",
551
+ name: "Nova Micro (EU)",
552
+ api: "bedrock-converse-stream",
553
+ provider: "amazon-bedrock",
554
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
555
+ reasoning: false,
556
+ input: ["text"],
557
+ cost: {
558
+ input: 0.04,
559
+ output: 0.16,
560
+ cacheRead: 0.01,
561
+ cacheWrite: 0.04,
562
+ },
563
+ contextWindow: 128000,
564
+ maxTokens: 10000,
565
+ },
566
+ "eu.amazon.nova-pro-v1:0": {
567
+ id: "eu.amazon.nova-pro-v1:0",
568
+ name: "Nova Pro (EU)",
569
+ api: "bedrock-converse-stream",
570
+ provider: "amazon-bedrock",
571
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
572
+ reasoning: false,
573
+ input: ["text", "image"],
574
+ cost: {
575
+ input: 0.92,
576
+ output: 3.68,
577
+ cacheRead: 0.23,
578
+ cacheWrite: 0.92,
579
+ },
580
+ contextWindow: 300000,
581
+ maxTokens: 10000,
582
+ },
447
583
  "eu.anthropic.claude-fable-5": {
448
584
  id: "eu.anthropic.claude-fable-5",
449
585
  name: "Claude Fable 5 (EU)",
@@ -614,6 +750,40 @@ export const MODELS = {
614
750
  contextWindow: 1000000,
615
751
  maxTokens: 128000,
616
752
  },
753
+ "eu.mistral.pixtral-large-2502-v1:0": {
754
+ id: "eu.mistral.pixtral-large-2502-v1:0",
755
+ name: "Pixtral Large (25.02) (EU)",
756
+ api: "bedrock-converse-stream",
757
+ provider: "amazon-bedrock",
758
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
759
+ reasoning: false,
760
+ input: ["text", "image"],
761
+ cost: {
762
+ input: 2,
763
+ output: 6,
764
+ cacheRead: 0,
765
+ cacheWrite: 0,
766
+ },
767
+ contextWindow: 128000,
768
+ maxTokens: 8192,
769
+ },
770
+ "global.amazon.nova-2-lite-v1:0": {
771
+ id: "global.amazon.nova-2-lite-v1:0",
772
+ name: "Nova 2 Lite (Global)",
773
+ api: "bedrock-converse-stream",
774
+ provider: "amazon-bedrock",
775
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
776
+ reasoning: true,
777
+ input: ["text", "image"],
778
+ cost: {
779
+ input: 0.3,
780
+ output: 2.5,
781
+ cacheRead: 0.075,
782
+ cacheWrite: 0.3,
783
+ },
784
+ contextWindow: 1000000,
785
+ maxTokens: 65535,
786
+ },
617
787
  "global.anthropic.claude-fable-5": {
618
788
  id: "global.anthropic.claude-fable-5",
619
789
  name: "Claude Fable 5 (Global)",
@@ -852,30 +1022,81 @@ export const MODELS = {
852
1022
  contextWindow: 1050000,
853
1023
  maxTokens: 128000,
854
1024
  },
855
- "google.gemma-3-27b-it": {
856
- id: "google.gemma-3-27b-it",
857
- name: "Google Gemma 3 27B Instruct",
1025
+ "global.openai.gpt-6-astra": {
1026
+ id: "global.openai.gpt-6-astra",
1027
+ name: "GPT-6 Astra (Global)",
858
1028
  api: "bedrock-converse-stream",
859
1029
  provider: "amazon-bedrock",
860
1030
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
861
- reasoning: false,
1031
+ reasoning: true,
862
1032
  input: ["text", "image"],
863
1033
  cost: {
864
- input: 0.12,
865
- output: 0.2,
1034
+ input: 10,
1035
+ output: 50,
1036
+ cacheRead: 1,
1037
+ cacheWrite: 12.5,
1038
+ },
1039
+ contextWindow: 1050000,
1040
+ maxTokens: 128000,
1041
+ },
1042
+ "global.xai.grok-4.6": {
1043
+ id: "global.xai.grok-4.6",
1044
+ name: "Grok 4.6 (Global)",
1045
+ api: "bedrock-converse-stream",
1046
+ provider: "amazon-bedrock",
1047
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1048
+ reasoning: true,
1049
+ input: ["text", "image"],
1050
+ cost: {
1051
+ input: 2,
1052
+ output: 6,
1053
+ cacheRead: 0.5,
1054
+ cacheWrite: 0,
1055
+ },
1056
+ contextWindow: 500000,
1057
+ maxTokens: 500000,
1058
+ },
1059
+ "google.gemma-4-26b-a4b": {
1060
+ id: "google.gemma-4-26b-a4b",
1061
+ name: "Gemma 4 26B A4B IT",
1062
+ api: "bedrock-converse-stream",
1063
+ provider: "amazon-bedrock",
1064
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1065
+ reasoning: true,
1066
+ input: ["text", "image"],
1067
+ cost: {
1068
+ input: 0.13,
1069
+ output: 0.4,
866
1070
  cacheRead: 0,
867
1071
  cacheWrite: 0,
868
1072
  },
869
- contextWindow: 202752,
870
- maxTokens: 8192,
1073
+ contextWindow: 262144,
1074
+ maxTokens: 32768,
871
1075
  },
872
- "google.gemma-3-4b-it": {
873
- id: "google.gemma-3-4b-it",
874
- name: "Gemma 3 4B IT",
1076
+ "google.gemma-4-31b": {
1077
+ id: "google.gemma-4-31b",
1078
+ name: "Gemma 4 31B IT",
875
1079
  api: "bedrock-converse-stream",
876
1080
  provider: "amazon-bedrock",
877
1081
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
878
- reasoning: false,
1082
+ reasoning: true,
1083
+ input: ["text", "image"],
1084
+ cost: {
1085
+ input: 0.14,
1086
+ output: 0.4,
1087
+ cacheRead: 0,
1088
+ cacheWrite: 0,
1089
+ },
1090
+ contextWindow: 262144,
1091
+ maxTokens: 32768,
1092
+ },
1093
+ "google.gemma-4-e2b": {
1094
+ id: "google.gemma-4-e2b",
1095
+ name: "Gemma 4 E2B IT",
1096
+ api: "bedrock-converse-stream",
1097
+ provider: "amazon-bedrock",
1098
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1099
+ reasoning: true,
879
1100
  input: ["text", "image"],
880
1101
  cost: {
881
1102
  input: 0.04,
@@ -883,8 +1104,59 @@ export const MODELS = {
883
1104
  cacheRead: 0,
884
1105
  cacheWrite: 0,
885
1106
  },
886
- contextWindow: 128000,
887
- maxTokens: 4096,
1107
+ contextWindow: 131072,
1108
+ maxTokens: 8192,
1109
+ },
1110
+ "in.openai.gpt-5.6-luna": {
1111
+ id: "in.openai.gpt-5.6-luna",
1112
+ name: "GPT-5.6 Luna (India)",
1113
+ api: "bedrock-converse-stream",
1114
+ provider: "amazon-bedrock",
1115
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1116
+ reasoning: true,
1117
+ input: ["text", "image"],
1118
+ cost: {
1119
+ input: 0.22,
1120
+ output: 1.32,
1121
+ cacheRead: 0.022,
1122
+ cacheWrite: 0.275,
1123
+ },
1124
+ contextWindow: 1050000,
1125
+ maxTokens: 128000,
1126
+ },
1127
+ "in.openai.gpt-5.6-terra": {
1128
+ id: "in.openai.gpt-5.6-terra",
1129
+ name: "GPT-5.6 Terra (India)",
1130
+ api: "bedrock-converse-stream",
1131
+ provider: "amazon-bedrock",
1132
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1133
+ reasoning: true,
1134
+ input: ["text", "image"],
1135
+ cost: {
1136
+ input: 2.2,
1137
+ output: 13.2,
1138
+ cacheRead: 0.22,
1139
+ cacheWrite: 2.75,
1140
+ },
1141
+ contextWindow: 1050000,
1142
+ maxTokens: 128000,
1143
+ },
1144
+ "jp.amazon.nova-2-lite-v1:0": {
1145
+ id: "jp.amazon.nova-2-lite-v1:0",
1146
+ name: "Nova 2 Lite (JP)",
1147
+ api: "bedrock-converse-stream",
1148
+ provider: "amazon-bedrock",
1149
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1150
+ reasoning: true,
1151
+ input: ["text", "image"],
1152
+ cost: {
1153
+ input: 0.396,
1154
+ output: 3.311,
1155
+ cacheRead: 0.099,
1156
+ cacheWrite: 0.396,
1157
+ },
1158
+ contextWindow: 1000000,
1159
+ maxTokens: 65535,
888
1160
  },
889
1161
  "jp.anthropic.claude-haiku-4-5-20251001-v1:0": {
890
1162
  id: "jp.anthropic.claude-haiku-4-5-20251001-v1:0",
@@ -1071,7 +1343,7 @@ export const MODELS = {
1071
1343
  cacheWrite: 0,
1072
1344
  },
1073
1345
  contextWindow: 1000000,
1074
- maxTokens: 16384,
1346
+ maxTokens: 8192,
1075
1347
  },
1076
1348
  "meta.llama4-scout-17b-instruct-v1:0": {
1077
1349
  id: "meta.llama4-scout-17b-instruct-v1:0",
@@ -1087,12 +1359,12 @@ export const MODELS = {
1087
1359
  cacheRead: 0,
1088
1360
  cacheWrite: 0,
1089
1361
  },
1090
- contextWindow: 3500000,
1091
- maxTokens: 16384,
1362
+ contextWindow: 10000000,
1363
+ maxTokens: 8192,
1092
1364
  },
1093
1365
  "minimax.minimax-m2": {
1094
1366
  id: "minimax.minimax-m2",
1095
- name: "MiniMax M2",
1367
+ name: "MiniMax-M2",
1096
1368
  api: "bedrock-converse-stream",
1097
1369
  provider: "amazon-bedrock",
1098
1370
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
@@ -1109,7 +1381,7 @@ export const MODELS = {
1109
1381
  },
1110
1382
  "minimax.minimax-m2.1": {
1111
1383
  id: "minimax.minimax-m2.1",
1112
- name: "MiniMax M2.1",
1384
+ name: "MiniMax-M2.1",
1113
1385
  api: "bedrock-converse-stream",
1114
1386
  provider: "amazon-bedrock",
1115
1387
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
@@ -1126,7 +1398,7 @@ export const MODELS = {
1126
1398
  },
1127
1399
  "minimax.minimax-m2.5": {
1128
1400
  id: "minimax.minimax-m2.5",
1129
- name: "MiniMax M2.5",
1401
+ name: "MiniMax-M2.5",
1130
1402
  api: "bedrock-converse-stream",
1131
1403
  provider: "amazon-bedrock",
1132
1404
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
@@ -1182,7 +1454,7 @@ export const MODELS = {
1182
1454
  provider: "amazon-bedrock",
1183
1455
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1184
1456
  reasoning: false,
1185
- input: ["text"],
1457
+ input: ["text", "image"],
1186
1458
  cost: {
1187
1459
  input: 0.2,
1188
1460
  output: 0.2,
@@ -1216,7 +1488,7 @@ export const MODELS = {
1216
1488
  provider: "amazon-bedrock",
1217
1489
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1218
1490
  reasoning: false,
1219
- input: ["text"],
1491
+ input: ["text", "image"],
1220
1492
  cost: {
1221
1493
  input: 0.15,
1222
1494
  output: 0.15,
@@ -1274,7 +1546,7 @@ export const MODELS = {
1274
1546
  cacheRead: 0,
1275
1547
  cacheWrite: 0,
1276
1548
  },
1277
- contextWindow: 128000,
1549
+ contextWindow: 32768,
1278
1550
  maxTokens: 4096,
1279
1551
  },
1280
1552
  "mistral.voxtral-small-24b-2507": {
@@ -1286,12 +1558,12 @@ export const MODELS = {
1286
1558
  reasoning: false,
1287
1559
  input: ["text"],
1288
1560
  cost: {
1289
- input: 0.15,
1290
- output: 0.35,
1561
+ input: 0.1,
1562
+ output: 0.3,
1291
1563
  cacheRead: 0,
1292
1564
  cacheWrite: 0,
1293
1565
  },
1294
- contextWindow: 32000,
1566
+ contextWindow: 32768,
1295
1567
  maxTokens: 8192,
1296
1568
  },
1297
1569
  "moonshot.kimi-k2-thinking": {
@@ -1326,7 +1598,7 @@ export const MODELS = {
1326
1598
  cacheWrite: 0,
1327
1599
  },
1328
1600
  contextWindow: 262143,
1329
- maxTokens: 16000,
1601
+ maxTokens: 16384,
1330
1602
  },
1331
1603
  "nvidia.nemotron-nano-12b-v2": {
1332
1604
  id: "nvidia.nemotron-nano-12b-v2",
@@ -1343,7 +1615,7 @@ export const MODELS = {
1343
1615
  cacheWrite: 0,
1344
1616
  },
1345
1617
  contextWindow: 128000,
1346
- maxTokens: 4096,
1618
+ maxTokens: 8192,
1347
1619
  },
1348
1620
  "nvidia.nemotron-nano-3-30b": {
1349
1621
  id: "nvidia.nemotron-nano-3-30b",
@@ -1359,8 +1631,8 @@ export const MODELS = {
1359
1631
  cacheRead: 0,
1360
1632
  cacheWrite: 0,
1361
1633
  },
1362
- contextWindow: 128000,
1363
- maxTokens: 4096,
1634
+ contextWindow: 262144,
1635
+ maxTokens: 8192,
1364
1636
  },
1365
1637
  "nvidia.nemotron-nano-9b-v2": {
1366
1638
  id: "nvidia.nemotron-nano-9b-v2",
@@ -1376,8 +1648,8 @@ export const MODELS = {
1376
1648
  cacheRead: 0,
1377
1649
  cacheWrite: 0,
1378
1650
  },
1379
- contextWindow: 128000,
1380
- maxTokens: 4096,
1651
+ contextWindow: 131072,
1652
+ maxTokens: 8192,
1381
1653
  },
1382
1654
  "nvidia.nemotron-super-3-120b": {
1383
1655
  id: "nvidia.nemotron-super-3-120b",
@@ -1481,6 +1753,23 @@ export const MODELS = {
1481
1753
  contextWindow: 1050000,
1482
1754
  maxTokens: 128000,
1483
1755
  },
1756
+ "openai.gpt-6-astra": {
1757
+ id: "openai.gpt-6-astra",
1758
+ name: "GPT-6 Astra",
1759
+ api: "bedrock-converse-stream",
1760
+ provider: "amazon-bedrock",
1761
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1762
+ reasoning: true,
1763
+ input: ["text", "image"],
1764
+ cost: {
1765
+ input: 11,
1766
+ output: 55,
1767
+ cacheRead: 1.1,
1768
+ cacheWrite: 13.75,
1769
+ },
1770
+ contextWindow: 1050000,
1771
+ maxTokens: 128000,
1772
+ },
1484
1773
  "openai.gpt-oss-120b": {
1485
1774
  id: "openai.gpt-oss-120b",
1486
1775
  name: "gpt-oss-120b",
@@ -1555,7 +1844,7 @@ export const MODELS = {
1555
1844
  api: "bedrock-converse-stream",
1556
1845
  provider: "amazon-bedrock",
1557
1846
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1558
- reasoning: false,
1847
+ reasoning: true,
1559
1848
  input: ["text"],
1560
1849
  cost: {
1561
1850
  input: 0.15,
@@ -1572,7 +1861,7 @@ export const MODELS = {
1572
1861
  api: "bedrock-converse-stream",
1573
1862
  provider: "amazon-bedrock",
1574
1863
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1575
- reasoning: false,
1864
+ reasoning: true,
1576
1865
  input: ["text"],
1577
1866
  cost: {
1578
1867
  input: 0.07,
@@ -1585,7 +1874,7 @@ export const MODELS = {
1585
1874
  },
1586
1875
  "qwen.qwen3-235b-a22b-2507-v1:0": {
1587
1876
  id: "qwen.qwen3-235b-a22b-2507-v1:0",
1588
- name: "Qwen3 235B A22B 2507",
1877
+ name: "Qwen3 235B-A22B Instruct 2507",
1589
1878
  api: "bedrock-converse-stream",
1590
1879
  provider: "amazon-bedrock",
1591
1880
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
@@ -1602,7 +1891,7 @@ export const MODELS = {
1602
1891
  },
1603
1892
  "qwen.qwen3-32b-v1:0": {
1604
1893
  id: "qwen.qwen3-32b-v1:0",
1605
- name: "Qwen3 32B (dense)",
1894
+ name: "Qwen3 32B",
1606
1895
  api: "bedrock-converse-stream",
1607
1896
  provider: "amazon-bedrock",
1608
1897
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
@@ -1614,12 +1903,12 @@ export const MODELS = {
1614
1903
  cacheRead: 0,
1615
1904
  cacheWrite: 0,
1616
1905
  },
1617
- contextWindow: 16384,
1906
+ contextWindow: 32768,
1618
1907
  maxTokens: 16384,
1619
1908
  },
1620
1909
  "qwen.qwen3-coder-30b-a3b-v1:0": {
1621
1910
  id: "qwen.qwen3-coder-30b-a3b-v1:0",
1622
- name: "Qwen3 Coder 30B A3B Instruct",
1911
+ name: "Qwen3-Coder 30B-A3B Instruct",
1623
1912
  api: "bedrock-converse-stream",
1624
1913
  provider: "amazon-bedrock",
1625
1914
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
@@ -1636,14 +1925,14 @@ export const MODELS = {
1636
1925
  },
1637
1926
  "qwen.qwen3-coder-480b-a35b-v1:0": {
1638
1927
  id: "qwen.qwen3-coder-480b-a35b-v1:0",
1639
- name: "Qwen3 Coder 480B A35B Instruct",
1928
+ name: "Qwen3-Coder 480B-A35B Instruct",
1640
1929
  api: "bedrock-converse-stream",
1641
1930
  provider: "amazon-bedrock",
1642
1931
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1643
1932
  reasoning: false,
1644
1933
  input: ["text"],
1645
1934
  cost: {
1646
- input: 0.22,
1935
+ input: 0.45,
1647
1936
  output: 1.8,
1648
1937
  cacheRead: 0,
1649
1938
  cacheWrite: 0,
@@ -1657,51 +1946,170 @@ export const MODELS = {
1657
1946
  api: "bedrock-converse-stream",
1658
1947
  provider: "amazon-bedrock",
1659
1948
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1660
- reasoning: true,
1949
+ reasoning: false,
1661
1950
  input: ["text"],
1662
1951
  cost: {
1663
- input: 0.22,
1664
- output: 1.8,
1952
+ input: 0.5,
1953
+ output: 1.2,
1665
1954
  cacheRead: 0,
1666
1955
  cacheWrite: 0,
1667
1956
  },
1668
- contextWindow: 131072,
1957
+ contextWindow: 262144,
1669
1958
  maxTokens: 65536,
1670
1959
  },
1671
1960
  "qwen.qwen3-next-80b-a3b": {
1672
1961
  id: "qwen.qwen3-next-80b-a3b",
1673
- name: "Qwen/Qwen3-Next-80B-A3B-Instruct",
1962
+ name: "Qwen3-Next 80B-A3B Instruct",
1674
1963
  api: "bedrock-converse-stream",
1675
1964
  provider: "amazon-bedrock",
1676
1965
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1677
1966
  reasoning: false,
1678
1967
  input: ["text"],
1679
1968
  cost: {
1680
- input: 0.14,
1681
- output: 1.4,
1969
+ input: 0.15,
1970
+ output: 1.2,
1682
1971
  cacheRead: 0,
1683
1972
  cacheWrite: 0,
1684
1973
  },
1685
- contextWindow: 262000,
1974
+ contextWindow: 262144,
1686
1975
  maxTokens: 262000,
1687
1976
  },
1688
1977
  "qwen.qwen3-vl-235b-a22b": {
1689
1978
  id: "qwen.qwen3-vl-235b-a22b",
1690
- name: "Qwen/Qwen3-VL-235B-A22B-Instruct",
1979
+ name: "Qwen3 VL 235B A22B Instruct",
1691
1980
  api: "bedrock-converse-stream",
1692
1981
  provider: "amazon-bedrock",
1693
1982
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1694
1983
  reasoning: false,
1695
1984
  input: ["text", "image"],
1696
1985
  cost: {
1697
- input: 0.3,
1698
- output: 1.5,
1986
+ input: 0.53,
1987
+ output: 2.66,
1699
1988
  cacheRead: 0,
1700
1989
  cacheWrite: 0,
1701
1990
  },
1702
- contextWindow: 262000,
1991
+ contextWindow: 262144,
1703
1992
  maxTokens: 262000,
1704
1993
  },
1994
+ "us-gov.openai.gpt-oss-120b-1:0": {
1995
+ id: "us-gov.openai.gpt-oss-120b-1:0",
1996
+ name: "gpt-oss-120b (GovCloud)",
1997
+ api: "bedrock-converse-stream",
1998
+ provider: "amazon-bedrock",
1999
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2000
+ reasoning: true,
2001
+ input: ["text"],
2002
+ cost: {
2003
+ input: 0.18,
2004
+ output: 0.72,
2005
+ cacheRead: 0,
2006
+ cacheWrite: 0,
2007
+ },
2008
+ contextWindow: 128000,
2009
+ maxTokens: 16384,
2010
+ },
2011
+ "us-gov.openai.gpt-oss-20b-1:0": {
2012
+ id: "us-gov.openai.gpt-oss-20b-1:0",
2013
+ name: "gpt-oss-20b (GovCloud)",
2014
+ api: "bedrock-converse-stream",
2015
+ provider: "amazon-bedrock",
2016
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2017
+ reasoning: true,
2018
+ input: ["text"],
2019
+ cost: {
2020
+ input: 0.084,
2021
+ output: 0.36,
2022
+ cacheRead: 0,
2023
+ cacheWrite: 0,
2024
+ },
2025
+ contextWindow: 128000,
2026
+ maxTokens: 16384,
2027
+ },
2028
+ "us.amazon.nova-2-lite-v1:0": {
2029
+ id: "us.amazon.nova-2-lite-v1:0",
2030
+ name: "Nova 2 Lite (US)",
2031
+ api: "bedrock-converse-stream",
2032
+ provider: "amazon-bedrock",
2033
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2034
+ reasoning: true,
2035
+ input: ["text", "image"],
2036
+ cost: {
2037
+ input: 0.33,
2038
+ output: 2.75,
2039
+ cacheRead: 0.0825,
2040
+ cacheWrite: 0.33,
2041
+ },
2042
+ contextWindow: 1000000,
2043
+ maxTokens: 65535,
2044
+ },
2045
+ "us.amazon.nova-lite-v1:0": {
2046
+ id: "us.amazon.nova-lite-v1:0",
2047
+ name: "Nova Lite (US)",
2048
+ api: "bedrock-converse-stream",
2049
+ provider: "amazon-bedrock",
2050
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2051
+ reasoning: false,
2052
+ input: ["text", "image"],
2053
+ cost: {
2054
+ input: 0.06,
2055
+ output: 0.24,
2056
+ cacheRead: 0.015,
2057
+ cacheWrite: 0.06,
2058
+ },
2059
+ contextWindow: 300000,
2060
+ maxTokens: 10000,
2061
+ },
2062
+ "us.amazon.nova-micro-v1:0": {
2063
+ id: "us.amazon.nova-micro-v1:0",
2064
+ name: "Nova Micro (US)",
2065
+ api: "bedrock-converse-stream",
2066
+ provider: "amazon-bedrock",
2067
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2068
+ reasoning: false,
2069
+ input: ["text"],
2070
+ cost: {
2071
+ input: 0.035,
2072
+ output: 0.14,
2073
+ cacheRead: 0.00875,
2074
+ cacheWrite: 0.035,
2075
+ },
2076
+ contextWindow: 128000,
2077
+ maxTokens: 10000,
2078
+ },
2079
+ "us.amazon.nova-premier-v1:0": {
2080
+ id: "us.amazon.nova-premier-v1:0",
2081
+ name: "Nova Premier (US)",
2082
+ api: "bedrock-converse-stream",
2083
+ provider: "amazon-bedrock",
2084
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2085
+ reasoning: false,
2086
+ input: ["text", "image"],
2087
+ cost: {
2088
+ input: 2.5,
2089
+ output: 12.5,
2090
+ cacheRead: 0.625,
2091
+ cacheWrite: 2.5,
2092
+ },
2093
+ contextWindow: 1000000,
2094
+ maxTokens: 10000,
2095
+ },
2096
+ "us.amazon.nova-pro-v1:0": {
2097
+ id: "us.amazon.nova-pro-v1:0",
2098
+ name: "Nova Pro (US)",
2099
+ api: "bedrock-converse-stream",
2100
+ provider: "amazon-bedrock",
2101
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2102
+ reasoning: false,
2103
+ input: ["text", "image"],
2104
+ cost: {
2105
+ input: 0.8,
2106
+ output: 3.2,
2107
+ cacheRead: 0.2,
2108
+ cacheWrite: 0.8,
2109
+ },
2110
+ contextWindow: 300000,
2111
+ maxTokens: 10000,
2112
+ },
1705
2113
  "us.anthropic.claude-fable-5": {
1706
2114
  id: "us.anthropic.claude-fable-5",
1707
2115
  name: "Claude Fable 5 (US)",
@@ -1886,76 +2294,246 @@ export const MODELS = {
1886
2294
  cacheRead: 0.3,
1887
2295
  cacheWrite: 3.75,
1888
2296
  },
1889
- contextWindow: 1000000,
1890
- maxTokens: 64000,
2297
+ contextWindow: 1000000,
2298
+ maxTokens: 64000,
2299
+ },
2300
+ "us.anthropic.claude-sonnet-5": {
2301
+ id: "us.anthropic.claude-sonnet-5",
2302
+ name: "Claude Sonnet 5 (US)",
2303
+ api: "bedrock-converse-stream",
2304
+ provider: "amazon-bedrock",
2305
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2306
+ reasoning: true,
2307
+ input: ["text", "image"],
2308
+ cost: {
2309
+ input: 2,
2310
+ output: 10,
2311
+ cacheRead: 0.2,
2312
+ cacheWrite: 2.5,
2313
+ },
2314
+ contextWindow: 1000000,
2315
+ maxTokens: 128000,
2316
+ },
2317
+ "us.meta.llama3-1-70b-instruct-v1:0": {
2318
+ id: "us.meta.llama3-1-70b-instruct-v1:0",
2319
+ name: "Llama 3.1 70B Instruct (US)",
2320
+ api: "bedrock-converse-stream",
2321
+ provider: "amazon-bedrock",
2322
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2323
+ reasoning: false,
2324
+ input: ["text"],
2325
+ cost: {
2326
+ input: 0.72,
2327
+ output: 0.72,
2328
+ cacheRead: 0,
2329
+ cacheWrite: 0,
2330
+ },
2331
+ contextWindow: 128000,
2332
+ maxTokens: 4096,
2333
+ },
2334
+ "us.meta.llama3-1-8b-instruct-v1:0": {
2335
+ id: "us.meta.llama3-1-8b-instruct-v1:0",
2336
+ name: "Llama 3.1 8B Instruct (US)",
2337
+ api: "bedrock-converse-stream",
2338
+ provider: "amazon-bedrock",
2339
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2340
+ reasoning: false,
2341
+ input: ["text"],
2342
+ cost: {
2343
+ input: 0.22,
2344
+ output: 0.22,
2345
+ cacheRead: 0,
2346
+ cacheWrite: 0,
2347
+ },
2348
+ contextWindow: 128000,
2349
+ maxTokens: 4096,
2350
+ },
2351
+ "us.meta.llama3-3-70b-instruct-v1:0": {
2352
+ id: "us.meta.llama3-3-70b-instruct-v1:0",
2353
+ name: "Llama 3.3 70B Instruct (US)",
2354
+ api: "bedrock-converse-stream",
2355
+ provider: "amazon-bedrock",
2356
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2357
+ reasoning: false,
2358
+ input: ["text"],
2359
+ cost: {
2360
+ input: 0.72,
2361
+ output: 0.72,
2362
+ cacheRead: 0,
2363
+ cacheWrite: 0,
2364
+ },
2365
+ contextWindow: 128000,
2366
+ maxTokens: 4096,
2367
+ },
2368
+ "us.meta.llama4-maverick-17b-instruct-v1:0": {
2369
+ id: "us.meta.llama4-maverick-17b-instruct-v1:0",
2370
+ name: "Llama 4 Maverick 17B Instruct (US)",
2371
+ api: "bedrock-converse-stream",
2372
+ provider: "amazon-bedrock",
2373
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2374
+ reasoning: false,
2375
+ input: ["text", "image"],
2376
+ cost: {
2377
+ input: 0.24,
2378
+ output: 0.97,
2379
+ cacheRead: 0,
2380
+ cacheWrite: 0,
2381
+ },
2382
+ contextWindow: 1000000,
2383
+ maxTokens: 8192,
2384
+ },
2385
+ "us.meta.llama4-scout-17b-instruct-v1:0": {
2386
+ id: "us.meta.llama4-scout-17b-instruct-v1:0",
2387
+ name: "Llama 4 Scout 17B Instruct (US)",
2388
+ api: "bedrock-converse-stream",
2389
+ provider: "amazon-bedrock",
2390
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2391
+ reasoning: false,
2392
+ input: ["text", "image"],
2393
+ cost: {
2394
+ input: 0.17,
2395
+ output: 0.66,
2396
+ cacheRead: 0,
2397
+ cacheWrite: 0,
2398
+ },
2399
+ contextWindow: 10000000,
2400
+ maxTokens: 8192,
2401
+ },
2402
+ "us.mistral.pixtral-large-2502-v1:0": {
2403
+ id: "us.mistral.pixtral-large-2502-v1:0",
2404
+ name: "Pixtral Large (25.02) (US)",
2405
+ api: "bedrock-converse-stream",
2406
+ provider: "amazon-bedrock",
2407
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2408
+ reasoning: false,
2409
+ input: ["text", "image"],
2410
+ cost: {
2411
+ input: 2,
2412
+ output: 6,
2413
+ cacheRead: 0,
2414
+ cacheWrite: 0,
2415
+ },
2416
+ contextWindow: 128000,
2417
+ maxTokens: 8192,
2418
+ },
2419
+ "us.openai.gpt-5.6-luna": {
2420
+ id: "us.openai.gpt-5.6-luna",
2421
+ name: "GPT-5.6 Luna (US)",
2422
+ api: "bedrock-converse-stream",
2423
+ provider: "amazon-bedrock",
2424
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2425
+ reasoning: true,
2426
+ input: ["text", "image"],
2427
+ cost: {
2428
+ input: 0.22,
2429
+ output: 1.32,
2430
+ cacheRead: 0.022,
2431
+ cacheWrite: 0.275,
2432
+ },
2433
+ contextWindow: 1050000,
2434
+ maxTokens: 128000,
2435
+ },
2436
+ "us.openai.gpt-5.6-sol": {
2437
+ id: "us.openai.gpt-5.6-sol",
2438
+ name: "GPT-5.6 Sol (US)",
2439
+ api: "bedrock-converse-stream",
2440
+ provider: "amazon-bedrock",
2441
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2442
+ reasoning: true,
2443
+ input: ["text", "image"],
2444
+ cost: {
2445
+ input: 4.4,
2446
+ output: 22,
2447
+ cacheRead: 0.44,
2448
+ cacheWrite: 5.5,
2449
+ },
2450
+ contextWindow: 1050000,
2451
+ maxTokens: 128000,
2452
+ },
2453
+ "us.openai.gpt-5.6-terra": {
2454
+ id: "us.openai.gpt-5.6-terra",
2455
+ name: "GPT-5.6 Terra (US)",
2456
+ api: "bedrock-converse-stream",
2457
+ provider: "amazon-bedrock",
2458
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2459
+ reasoning: true,
2460
+ input: ["text", "image"],
2461
+ cost: {
2462
+ input: 2.2,
2463
+ output: 13.2,
2464
+ cacheRead: 0.22,
2465
+ cacheWrite: 2.75,
2466
+ },
2467
+ contextWindow: 1050000,
2468
+ maxTokens: 128000,
1891
2469
  },
1892
- "us.anthropic.claude-sonnet-5": {
1893
- id: "us.anthropic.claude-sonnet-5",
1894
- name: "Claude Sonnet 5 (US)",
2470
+ "us.openai.gpt-6-astra": {
2471
+ id: "us.openai.gpt-6-astra",
2472
+ name: "GPT-6 Astra (US)",
1895
2473
  api: "bedrock-converse-stream",
1896
2474
  provider: "amazon-bedrock",
1897
2475
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1898
2476
  reasoning: true,
1899
2477
  input: ["text", "image"],
1900
2478
  cost: {
1901
- input: 2,
1902
- output: 10,
1903
- cacheRead: 0.2,
1904
- cacheWrite: 2.5,
2479
+ input: 11,
2480
+ output: 55,
2481
+ cacheRead: 1.1,
2482
+ cacheWrite: 13.75,
1905
2483
  },
1906
- contextWindow: 1000000,
2484
+ contextWindow: 1050000,
1907
2485
  maxTokens: 128000,
1908
2486
  },
1909
- "us.deepseek.r1-v1:0": {
1910
- id: "us.deepseek.r1-v1:0",
1911
- name: "DeepSeek-R1 (US)",
2487
+ "us.writer.palmyra-x4-v1:0": {
2488
+ id: "us.writer.palmyra-x4-v1:0",
2489
+ name: "Palmyra X4 (US)",
1912
2490
  api: "bedrock-converse-stream",
1913
2491
  provider: "amazon-bedrock",
1914
2492
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1915
2493
  reasoning: true,
1916
2494
  input: ["text"],
1917
2495
  cost: {
1918
- input: 1.35,
1919
- output: 5.4,
2496
+ input: 2.5,
2497
+ output: 10,
1920
2498
  cacheRead: 0,
1921
2499
  cacheWrite: 0,
1922
2500
  },
1923
- contextWindow: 128000,
1924
- maxTokens: 32768,
2501
+ contextWindow: 122880,
2502
+ maxTokens: 8192,
1925
2503
  },
1926
- "us.meta.llama4-maverick-17b-instruct-v1:0": {
1927
- id: "us.meta.llama4-maverick-17b-instruct-v1:0",
1928
- name: "Llama 4 Maverick 17B Instruct (US)",
2504
+ "us.writer.palmyra-x5-v1:0": {
2505
+ id: "us.writer.palmyra-x5-v1:0",
2506
+ name: "Palmyra X5 (US)",
1929
2507
  api: "bedrock-converse-stream",
1930
2508
  provider: "amazon-bedrock",
1931
2509
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1932
- reasoning: false,
1933
- input: ["text", "image"],
2510
+ reasoning: true,
2511
+ input: ["text"],
1934
2512
  cost: {
1935
- input: 0.24,
1936
- output: 0.97,
2513
+ input: 0.6,
2514
+ output: 6,
1937
2515
  cacheRead: 0,
1938
2516
  cacheWrite: 0,
1939
2517
  },
1940
- contextWindow: 1000000,
1941
- maxTokens: 16384,
2518
+ contextWindow: 1040000,
2519
+ maxTokens: 8192,
1942
2520
  },
1943
- "us.meta.llama4-scout-17b-instruct-v1:0": {
1944
- id: "us.meta.llama4-scout-17b-instruct-v1:0",
1945
- name: "Llama 4 Scout 17B Instruct (US)",
2521
+ "us.xai.grok-4.6": {
2522
+ id: "us.xai.grok-4.6",
2523
+ name: "Grok 4.6 (US)",
1946
2524
  api: "bedrock-converse-stream",
1947
2525
  provider: "amazon-bedrock",
1948
2526
  baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1949
- reasoning: false,
2527
+ reasoning: true,
1950
2528
  input: ["text", "image"],
1951
2529
  cost: {
1952
- input: 0.17,
1953
- output: 0.66,
1954
- cacheRead: 0,
2530
+ input: 2.2,
2531
+ output: 6.6,
2532
+ cacheRead: 0.55,
1955
2533
  cacheWrite: 0,
1956
2534
  },
1957
- contextWindow: 3500000,
1958
- maxTokens: 16384,
2535
+ contextWindow: 500000,
2536
+ maxTokens: 500000,
1959
2537
  },
1960
2538
  "writer.palmyra-x4-v1:0": {
1961
2539
  id: "writer.palmyra-x4-v1:0",
@@ -2074,7 +2652,7 @@ export const MODELS = {
2074
2652
  cacheWrite: 0,
2075
2653
  },
2076
2654
  contextWindow: 202752,
2077
- maxTokens: 101376,
2655
+ maxTokens: 131072,
2078
2656
  },
2079
2657
  },
2080
2658
  "anthropic": {
@@ -3034,39 +3612,39 @@ export const MODELS = {
3034
3612
  },
3035
3613
  },
3036
3614
  "cerebras": {
3037
- "gemma-4-31b": {
3038
- id: "gemma-4-31b",
3039
- name: "Gemma 4 31B IT",
3615
+ "gpt-oss-120b": {
3616
+ id: "gpt-oss-120b",
3617
+ name: "GPT OSS 120B",
3040
3618
  api: "openai-completions",
3041
3619
  provider: "cerebras",
3042
3620
  baseUrl: "https://api.cerebras.ai/v1",
3043
3621
  reasoning: true,
3044
- input: ["text", "image"],
3622
+ input: ["text"],
3045
3623
  cost: {
3046
- input: 0.99,
3047
- output: 1.49,
3624
+ input: 0.35,
3625
+ output: 0.75,
3048
3626
  cacheRead: 0,
3049
3627
  cacheWrite: 0,
3050
3628
  },
3051
3629
  contextWindow: 131072,
3052
3630
  maxTokens: 40960,
3053
3631
  },
3054
- "gpt-oss-120b": {
3055
- id: "gpt-oss-120b",
3056
- name: "GPT OSS 120B",
3632
+ "qwen-3.8-27b": {
3633
+ id: "qwen-3.8-27b",
3634
+ name: "Qwen3.8 27B",
3057
3635
  api: "openai-completions",
3058
3636
  provider: "cerebras",
3059
3637
  baseUrl: "https://api.cerebras.ai/v1",
3060
3638
  reasoning: true,
3061
- input: ["text"],
3639
+ input: ["text", "image"],
3062
3640
  cost: {
3063
- input: 0.35,
3064
- output: 0.75,
3641
+ input: 0.99,
3642
+ output: 1.49,
3065
3643
  cacheRead: 0,
3066
3644
  cacheWrite: 0,
3067
3645
  },
3068
- contextWindow: 131072,
3069
- maxTokens: 40960,
3646
+ contextWindow: 65536,
3647
+ maxTokens: 32768,
3070
3648
  },
3071
3649
  },
3072
3650
  "github-copilot": {
@@ -3459,11 +4037,10 @@ export const MODELS = {
3459
4037
  "gpt-6-astra": {
3460
4038
  id: "gpt-6-astra",
3461
4039
  name: "GPT-6 Astra",
3462
- api: "openai-completions",
4040
+ api: "openai-responses",
3463
4041
  provider: "github-copilot",
3464
4042
  baseUrl: "https://api.individual.githubcopilot.com",
3465
4043
  headers: { "User-Agent": "GitHubCopilotChat/0.35.0", "Editor-Version": "vscode/1.107.0", "Editor-Plugin-Version": "copilot-chat/0.35.0", "Copilot-Integration-Id": "vscode-chat" },
3466
- compat: { "supportsStore": false, "supportsDeveloperRole": false, "supportsReasoningEffort": false },
3467
4044
  reasoning: true,
3468
4045
  input: ["text", "image"],
3469
4046
  cost: {
@@ -5291,6 +5868,78 @@ export const MODELS = {
5291
5868
  contextWindow: 1000000,
5292
5869
  maxTokens: 384000,
5293
5870
  },
5871
+ "deepseek-ai/DeepSeek-V4.1-Flash": {
5872
+ id: "deepseek-ai/DeepSeek-V4.1-Flash",
5873
+ name: "DeepSeek V4.1 Flash",
5874
+ api: "openai-completions",
5875
+ provider: "huggingface",
5876
+ baseUrl: "https://router.huggingface.co/v1",
5877
+ compat: { "supportsDeveloperRole": false },
5878
+ reasoning: true,
5879
+ input: ["text", "image"],
5880
+ cost: {
5881
+ input: 0.3,
5882
+ output: 1.2,
5883
+ cacheRead: 0,
5884
+ cacheWrite: 0,
5885
+ },
5886
+ contextWindow: 1048576,
5887
+ maxTokens: 384000,
5888
+ },
5889
+ "google/gemma-3-12b-it": {
5890
+ id: "google/gemma-3-12b-it",
5891
+ name: "Gemma 3 12B IT",
5892
+ api: "openai-completions",
5893
+ provider: "huggingface",
5894
+ baseUrl: "https://router.huggingface.co/v1",
5895
+ compat: { "supportsDeveloperRole": false },
5896
+ reasoning: false,
5897
+ input: ["text", "image"],
5898
+ cost: {
5899
+ input: 0.05,
5900
+ output: 0.15,
5901
+ cacheRead: 0,
5902
+ cacheWrite: 0,
5903
+ },
5904
+ contextWindow: 131072,
5905
+ maxTokens: 131072,
5906
+ },
5907
+ "google/gemma-3-27b-it": {
5908
+ id: "google/gemma-3-27b-it",
5909
+ name: "Gemma 3 27B IT",
5910
+ api: "openai-completions",
5911
+ provider: "huggingface",
5912
+ baseUrl: "https://router.huggingface.co/v1",
5913
+ compat: { "supportsDeveloperRole": false },
5914
+ reasoning: false,
5915
+ input: ["text", "image"],
5916
+ cost: {
5917
+ input: 0.08,
5918
+ output: 0.16,
5919
+ cacheRead: 0,
5920
+ cacheWrite: 0,
5921
+ },
5922
+ contextWindow: 131072,
5923
+ maxTokens: 131072,
5924
+ },
5925
+ "google/gemma-3-4b-it": {
5926
+ id: "google/gemma-3-4b-it",
5927
+ name: "Gemma 3 4B IT",
5928
+ api: "openai-completions",
5929
+ provider: "huggingface",
5930
+ baseUrl: "https://router.huggingface.co/v1",
5931
+ compat: { "supportsDeveloperRole": false },
5932
+ reasoning: false,
5933
+ input: ["text", "image"],
5934
+ cost: {
5935
+ input: 0.05,
5936
+ output: 0.1,
5937
+ cacheRead: 0,
5938
+ cacheWrite: 0,
5939
+ },
5940
+ contextWindow: 131072,
5941
+ maxTokens: 131072,
5942
+ },
5294
5943
  "google/gemma-4-26B-A4B-it": {
5295
5944
  id: "google/gemma-4-26B-A4B-it",
5296
5945
  name: "Gemma 4 26B A4B IT",
@@ -7232,186 +7881,84 @@ export const MODELS = {
7232
7881
  reasoning: true,
7233
7882
  input: ["text", "image"],
7234
7883
  cost: {
7235
- input: 150,
7236
- output: 600,
7237
- cacheRead: 0,
7238
- cacheWrite: 0,
7239
- },
7240
- contextWindow: 200000,
7241
- maxTokens: 100000,
7242
- },
7243
- "o3": {
7244
- id: "o3",
7245
- name: "o3",
7246
- api: "openai-responses",
7247
- provider: "openai",
7248
- baseUrl: "https://api.openai.com/v1",
7249
- reasoning: true,
7250
- input: ["text", "image"],
7251
- cost: {
7252
- input: 2,
7253
- output: 8,
7254
- cacheRead: 0.5,
7255
- cacheWrite: 0,
7256
- },
7257
- contextWindow: 200000,
7258
- maxTokens: 100000,
7259
- },
7260
- "o3-mini": {
7261
- id: "o3-mini",
7262
- name: "o3-mini",
7263
- api: "openai-responses",
7264
- provider: "openai",
7265
- baseUrl: "https://api.openai.com/v1",
7266
- reasoning: true,
7267
- input: ["text"],
7268
- cost: {
7269
- input: 1.1,
7270
- output: 4.4,
7271
- cacheRead: 0.55,
7272
- cacheWrite: 0,
7273
- },
7274
- contextWindow: 200000,
7275
- maxTokens: 100000,
7276
- },
7277
- "o3-pro": {
7278
- id: "o3-pro",
7279
- name: "o3-pro",
7280
- api: "openai-responses",
7281
- provider: "openai",
7282
- baseUrl: "https://api.openai.com/v1",
7283
- reasoning: true,
7284
- input: ["text", "image"],
7285
- cost: {
7286
- input: 20,
7287
- output: 80,
7288
- cacheRead: 0,
7289
- cacheWrite: 0,
7290
- },
7291
- contextWindow: 200000,
7292
- maxTokens: 100000,
7293
- },
7294
- "o4-mini": {
7295
- id: "o4-mini",
7296
- name: "o4-mini",
7297
- api: "openai-responses",
7298
- provider: "openai",
7299
- baseUrl: "https://api.openai.com/v1",
7300
- reasoning: true,
7301
- input: ["text", "image"],
7302
- cost: {
7303
- input: 1.1,
7304
- output: 4.4,
7305
- cacheRead: 0.275,
7306
- cacheWrite: 0,
7307
- },
7308
- contextWindow: 200000,
7309
- maxTokens: 100000,
7310
- },
7311
- },
7312
- "openai-codex": {
7313
- "gpt-5.1": {
7314
- id: "gpt-5.1",
7315
- name: "GPT-5.1",
7316
- api: "openai-codex-responses",
7317
- provider: "openai-codex",
7318
- baseUrl: "https://chatgpt.com/backend-api",
7319
- reasoning: true,
7320
- input: ["text", "image"],
7321
- cost: {
7322
- input: 1.25,
7323
- output: 10,
7324
- cacheRead: 0.125,
7325
- cacheWrite: 0,
7326
- },
7327
- contextWindow: 272000,
7328
- maxTokens: 128000,
7329
- },
7330
- "gpt-5.1-codex-max": {
7331
- id: "gpt-5.1-codex-max",
7332
- name: "GPT-5.1 Codex Max",
7333
- api: "openai-codex-responses",
7334
- provider: "openai-codex",
7335
- baseUrl: "https://chatgpt.com/backend-api",
7336
- reasoning: true,
7337
- input: ["text", "image"],
7338
- cost: {
7339
- input: 1.25,
7340
- output: 10,
7341
- cacheRead: 0.125,
7884
+ input: 150,
7885
+ output: 600,
7886
+ cacheRead: 0,
7342
7887
  cacheWrite: 0,
7343
7888
  },
7344
- contextWindow: 272000,
7345
- maxTokens: 128000,
7889
+ contextWindow: 200000,
7890
+ maxTokens: 100000,
7346
7891
  },
7347
- "gpt-5.1-codex-mini": {
7348
- id: "gpt-5.1-codex-mini",
7349
- name: "GPT-5.1 Codex Mini",
7350
- api: "openai-codex-responses",
7351
- provider: "openai-codex",
7352
- baseUrl: "https://chatgpt.com/backend-api",
7892
+ "o3": {
7893
+ id: "o3",
7894
+ name: "o3",
7895
+ api: "openai-responses",
7896
+ provider: "openai",
7897
+ baseUrl: "https://api.openai.com/v1",
7353
7898
  reasoning: true,
7354
7899
  input: ["text", "image"],
7355
7900
  cost: {
7356
- input: 0.25,
7357
- output: 2,
7358
- cacheRead: 0.025,
7901
+ input: 2,
7902
+ output: 8,
7903
+ cacheRead: 0.5,
7359
7904
  cacheWrite: 0,
7360
7905
  },
7361
- contextWindow: 272000,
7362
- maxTokens: 128000,
7906
+ contextWindow: 200000,
7907
+ maxTokens: 100000,
7363
7908
  },
7364
- "gpt-5.3-codex": {
7365
- id: "gpt-5.3-codex",
7366
- name: "GPT-5.3 Codex",
7367
- api: "openai-codex-responses",
7368
- provider: "openai-codex",
7369
- baseUrl: "https://chatgpt.com/backend-api",
7909
+ "o3-mini": {
7910
+ id: "o3-mini",
7911
+ name: "o3-mini",
7912
+ api: "openai-responses",
7913
+ provider: "openai",
7914
+ baseUrl: "https://api.openai.com/v1",
7370
7915
  reasoning: true,
7371
- input: ["text", "image"],
7916
+ input: ["text"],
7372
7917
  cost: {
7373
- input: 1.75,
7374
- output: 14,
7375
- cacheRead: 0.175,
7918
+ input: 1.1,
7919
+ output: 4.4,
7920
+ cacheRead: 0.55,
7376
7921
  cacheWrite: 0,
7377
7922
  },
7378
- contextWindow: 272000,
7379
- maxTokens: 128000,
7923
+ contextWindow: 200000,
7924
+ maxTokens: 100000,
7380
7925
  },
7381
- "gpt-5.3-codex-spark": {
7382
- id: "gpt-5.3-codex-spark",
7383
- name: "GPT-5.3 Codex Spark",
7384
- api: "openai-codex-responses",
7385
- provider: "openai-codex",
7386
- baseUrl: "https://chatgpt.com/backend-api",
7926
+ "o3-pro": {
7927
+ id: "o3-pro",
7928
+ name: "o3-pro",
7929
+ api: "openai-responses",
7930
+ provider: "openai",
7931
+ baseUrl: "https://api.openai.com/v1",
7387
7932
  reasoning: true,
7388
- input: ["text"],
7933
+ input: ["text", "image"],
7389
7934
  cost: {
7390
- input: 0,
7391
- output: 0,
7935
+ input: 20,
7936
+ output: 80,
7392
7937
  cacheRead: 0,
7393
7938
  cacheWrite: 0,
7394
7939
  },
7395
- contextWindow: 128000,
7396
- maxTokens: 128000,
7940
+ contextWindow: 200000,
7941
+ maxTokens: 100000,
7397
7942
  },
7398
- "gpt-5.4": {
7399
- id: "gpt-5.4",
7400
- name: "GPT-5.4",
7401
- api: "openai-codex-responses",
7402
- provider: "openai-codex",
7403
- baseUrl: "https://chatgpt.com/backend-api",
7943
+ "o4-mini": {
7944
+ id: "o4-mini",
7945
+ name: "o4-mini",
7946
+ api: "openai-responses",
7947
+ provider: "openai",
7948
+ baseUrl: "https://api.openai.com/v1",
7404
7949
  reasoning: true,
7405
7950
  input: ["text", "image"],
7406
7951
  cost: {
7407
- input: 2.5,
7408
- output: 15,
7409
- cacheRead: 0.25,
7952
+ input: 1.1,
7953
+ output: 4.4,
7954
+ cacheRead: 0.275,
7410
7955
  cacheWrite: 0,
7411
7956
  },
7412
- contextWindow: 272000,
7413
- maxTokens: 128000,
7957
+ contextWindow: 200000,
7958
+ maxTokens: 100000,
7414
7959
  },
7960
+ },
7961
+ "openai-codex": {
7415
7962
  "gpt-5.4-mini": {
7416
7963
  id: "gpt-5.4-mini",
7417
7964
  name: "GPT-5.4 Mini",
@@ -7497,6 +8044,23 @@ export const MODELS = {
7497
8044
  contextWindow: 372000,
7498
8045
  maxTokens: 128000,
7499
8046
  },
8047
+ "gpt-6-astra": {
8048
+ id: "gpt-6-astra",
8049
+ name: "GPT-6 Astra",
8050
+ api: "openai-codex-responses",
8051
+ provider: "openai-codex",
8052
+ baseUrl: "https://chatgpt.com/backend-api",
8053
+ reasoning: true,
8054
+ input: ["text", "image"],
8055
+ cost: {
8056
+ input: 10,
8057
+ output: 50,
8058
+ cacheRead: 1,
8059
+ cacheWrite: 12.5,
8060
+ },
8061
+ contextWindow: 272000,
8062
+ maxTokens: 128000,
8063
+ },
7500
8064
  },
7501
8065
  "opencode": {
7502
8066
  "big-pickle": {
@@ -8666,9 +9230,9 @@ export const MODELS = {
8666
9230
  reasoning: true,
8667
9231
  input: ["text"],
8668
9232
  cost: {
8669
- input: 0.22,
8670
- output: 0.66,
8671
- cacheRead: 0.007,
9233
+ input: 0.15,
9234
+ output: 0.6,
9235
+ cacheRead: 0.003,
8672
9236
  cacheWrite: 0,
8673
9237
  },
8674
9238
  contextWindow: 1000000,
@@ -8683,9 +9247,9 @@ export const MODELS = {
8683
9247
  reasoning: true,
8684
9248
  input: ["text", "image"],
8685
9249
  cost: {
8686
- input: 0.22,
8687
- output: 0.66,
8688
- cacheRead: 0.007,
9250
+ input: 0.15,
9251
+ output: 0.6,
9252
+ cacheRead: 0.003,
8689
9253
  cacheWrite: 0,
8690
9254
  },
8691
9255
  contextWindow: 1000000,
@@ -8708,6 +9272,23 @@ export const MODELS = {
8708
9272
  contextWindow: 1000000,
8709
9273
  maxTokens: 384000,
8710
9274
  },
9275
+ "deepseek-v4.1-flash": {
9276
+ id: "deepseek-v4.1-flash",
9277
+ name: "DeepSeek V4.1 Flash",
9278
+ api: "openai-completions",
9279
+ provider: "opencode-go",
9280
+ baseUrl: "https://opencode.ai/zen/go/v1",
9281
+ reasoning: true,
9282
+ input: ["text", "image"],
9283
+ cost: {
9284
+ input: 0.15,
9285
+ output: 0.6,
9286
+ cacheRead: 0.003,
9287
+ cacheWrite: 0,
9288
+ },
9289
+ contextWindow: 1000000,
9290
+ maxTokens: 384000,
9291
+ },
8711
9292
  "glm-5.1": {
8712
9293
  id: "glm-5.1",
8713
9294
  name: "GLM-5.1",
@@ -8761,16 +9342,16 @@ export const MODELS = {
8761
9342
  },
8762
9343
  "glm-5.3-flash": {
8763
9344
  id: "glm-5.3-flash",
8764
- name: "GLM-5.3-Flash (2x usage)",
9345
+ name: "GLM-5.3-Flash",
8765
9346
  api: "openai-completions",
8766
9347
  provider: "opencode-go",
8767
9348
  baseUrl: "https://opencode.ai/zen/go/v1",
8768
9349
  reasoning: true,
8769
9350
  input: ["text", "image"],
8770
9351
  cost: {
8771
- input: 0.075,
8772
- output: 0.25,
8773
- cacheRead: 0.015,
9352
+ input: 0.15,
9353
+ output: 0.5,
9354
+ cacheRead: 0.03,
8774
9355
  cacheWrite: 0,
8775
9356
  },
8776
9357
  contextWindow: 1000000,
@@ -9014,23 +9595,6 @@ export const MODELS = {
9014
9595
  contextWindow: 1048576,
9015
9596
  maxTokens: 131072,
9016
9597
  },
9017
- "omen-alpha": {
9018
- id: "omen-alpha",
9019
- name: "Omen Alpha",
9020
- api: "openai-completions",
9021
- provider: "opencode-go",
9022
- baseUrl: "https://opencode.ai/zen/go/v1",
9023
- reasoning: true,
9024
- input: ["text", "image"],
9025
- cost: {
9026
- input: 0.2,
9027
- output: 0.66,
9028
- cacheRead: 0.04,
9029
- cacheWrite: 0,
9030
- },
9031
- contextWindow: 500000,
9032
- maxTokens: 128000,
9033
- },
9034
9598
  "qwen3.6-plus": {
9035
9599
  id: "qwen3.6-plus",
9036
9600
  name: "Qwen3.6 Plus",
@@ -9909,13 +10473,13 @@ export const MODELS = {
9909
10473
  reasoning: false,
9910
10474
  input: ["text"],
9911
10475
  cost: {
9912
- input: 0.32,
9913
- output: 0.8899999999999999,
10476
+ input: 0.2574,
10477
+ output: 1.0287,
9914
10478
  cacheRead: 0,
9915
10479
  cacheWrite: 0,
9916
10480
  },
9917
10481
  contextWindow: 163840,
9918
- maxTokens: 16384,
10482
+ maxTokens: 16000,
9919
10483
  },
9920
10484
  "deepseek/deepseek-chat-v3-0324": {
9921
10485
  id: "deepseek/deepseek-chat-v3-0324",
@@ -9943,13 +10507,13 @@ export const MODELS = {
9943
10507
  reasoning: true,
9944
10508
  input: ["text"],
9945
10509
  cost: {
9946
- input: 0.55,
9947
- output: 1.6500000000000001,
9948
- cacheRead: 0.55,
10510
+ input: 0.25,
10511
+ output: 0.95,
10512
+ cacheRead: 0.13,
9949
10513
  cacheWrite: 0,
9950
10514
  },
9951
10515
  contextWindow: 163840,
9952
- maxTokens: 144900,
10516
+ maxTokens: 32768,
9953
10517
  },
9954
10518
  "deepseek/deepseek-r1": {
9955
10519
  id: "deepseek/deepseek-r1",
@@ -10045,9 +10609,9 @@ export const MODELS = {
10045
10609
  reasoning: true,
10046
10610
  input: ["text"],
10047
10611
  cost: {
10048
- input: 0.08078,
10049
- output: 0.16156,
10050
- cacheRead: 0.016156,
10612
+ input: 0.04676,
10613
+ output: 0.09352,
10614
+ cacheRead: 0.009352,
10051
10615
  cacheWrite: 0,
10052
10616
  },
10053
10617
  contextWindow: 1048576,
@@ -10062,13 +10626,13 @@ export const MODELS = {
10062
10626
  reasoning: true,
10063
10627
  input: ["text"],
10064
10628
  cost: {
10065
- input: 0.049980000000000004,
10066
- output: 0.09996000000000001,
10067
- cacheRead: 0.009996000000000001,
10629
+ input: 0.04,
10630
+ output: 0.08,
10631
+ cacheRead: 0.008,
10068
10632
  cacheWrite: 0,
10069
10633
  },
10070
10634
  contextWindow: 1310720,
10071
- maxTokens: 131072,
10635
+ maxTokens: 943718,
10072
10636
  },
10073
10637
  "deepseek/deepseek-v4-flash-0731:batch": {
10074
10638
  id: "deepseek/deepseek-v4-flash-0731:batch",
@@ -10079,9 +10643,9 @@ export const MODELS = {
10079
10643
  reasoning: true,
10080
10644
  input: ["text"],
10081
10645
  cost: {
10082
- input: 0.14,
10083
- output: 0.28,
10084
- cacheRead: 0.03,
10646
+ input: 0.11,
10647
+ output: 0.33,
10648
+ cacheRead: 0.0035,
10085
10649
  cacheWrite: 0,
10086
10650
  },
10087
10651
  contextWindow: 1048576,
@@ -10102,7 +10666,24 @@ export const MODELS = {
10102
10666
  cacheWrite: 0,
10103
10667
  },
10104
10668
  contextWindow: 1048576,
10105
- maxTokens: 384000,
10669
+ maxTokens: 943718,
10670
+ },
10671
+ "deepseek/deepseek-v4-flash-vision-exp:batch": {
10672
+ id: "deepseek/deepseek-v4-flash-vision-exp:batch",
10673
+ name: "DeepSeek: DeepSeek V4 Flash Vision Exp (batch)",
10674
+ api: "openai-completions",
10675
+ provider: "openrouter",
10676
+ baseUrl: "https://openrouter.ai/api/v1",
10677
+ reasoning: true,
10678
+ input: ["text", "image"],
10679
+ cost: {
10680
+ input: 0.11,
10681
+ output: 0.33,
10682
+ cacheRead: 0.0035,
10683
+ cacheWrite: 0,
10684
+ },
10685
+ contextWindow: 1048576,
10686
+ maxTokens: 943718,
10106
10687
  },
10107
10688
  "deepseek/deepseek-v4-pro": {
10108
10689
  id: "deepseek/deepseek-v4-pro",
@@ -10113,13 +10694,13 @@ export const MODELS = {
10113
10694
  reasoning: true,
10114
10695
  input: ["text"],
10115
10696
  cost: {
10116
- input: 0.69774,
10117
- output: 1.39548,
10118
- cacheRead: 0.058145,
10697
+ input: 1.5999999999999999,
10698
+ output: 3.1999999999999997,
10699
+ cacheRead: 0.135,
10119
10700
  cacheWrite: 0,
10120
10701
  },
10121
10702
  contextWindow: 1048576,
10122
- maxTokens: 384000,
10703
+ maxTokens: 393216,
10123
10704
  },
10124
10705
  "deepseek/deepseek-v4-pro-0813": {
10125
10706
  id: "deepseek/deepseek-v4-pro-0813",
@@ -10130,9 +10711,9 @@ export const MODELS = {
10130
10711
  reasoning: true,
10131
10712
  input: ["text"],
10132
10713
  cost: {
10133
- input: 1.12068,
10134
- output: 3.36204,
10135
- cacheRead: 0.037356,
10714
+ input: 0.57948,
10715
+ output: 1.73844,
10716
+ cacheRead: 0.019316,
10136
10717
  cacheWrite: 0,
10137
10718
  },
10138
10719
  contextWindow: 1048576,
@@ -10147,14 +10728,31 @@ export const MODELS = {
10147
10728
  reasoning: true,
10148
10729
  input: ["text"],
10149
10730
  cost: {
10150
- input: 1.32,
10151
- output: 3.9600000000000004,
10152
- cacheRead: 0.13,
10731
+ input: 0.66,
10732
+ output: 1.9800000000000002,
10733
+ cacheRead: 0.022,
10153
10734
  cacheWrite: 0,
10154
10735
  },
10155
10736
  contextWindow: 1048576,
10156
10737
  maxTokens: 943718,
10157
10738
  },
10739
+ "deepseek/deepseek-v4.1-flash": {
10740
+ id: "deepseek/deepseek-v4.1-flash",
10741
+ name: "DeepSeek: DeepSeek V4.1 Flash",
10742
+ api: "openai-completions",
10743
+ provider: "openrouter",
10744
+ baseUrl: "https://openrouter.ai/api/v1",
10745
+ reasoning: true,
10746
+ input: ["text", "image"],
10747
+ cost: {
10748
+ input: 0.15,
10749
+ output: 0.6,
10750
+ cacheRead: 0.003,
10751
+ cacheWrite: 0,
10752
+ },
10753
+ contextWindow: 1048576,
10754
+ maxTokens: 384000,
10755
+ },
10158
10756
  "dots-studio/dots-3-note-preview:free": {
10159
10757
  id: "dots-studio/dots-3-note-preview:free",
10160
10758
  name: "Dots Studio: Dots3-Note Preview (free)",
@@ -10674,13 +11272,13 @@ export const MODELS = {
10674
11272
  reasoning: true,
10675
11273
  input: ["text", "image"],
10676
11274
  cost: {
10677
- input: 0.07,
10678
- output: 0.33999999999999997,
10679
- cacheRead: 0,
11275
+ input: 0.09,
11276
+ output: 0.3,
11277
+ cacheRead: 0.049999999999999996,
10680
11278
  cacheWrite: 0,
10681
11279
  },
10682
11280
  contextWindow: 262144,
10683
- maxTokens: 16384,
11281
+ maxTokens: 235929,
10684
11282
  },
10685
11283
  "google/gemma-4-26b-a4b-it:free": {
10686
11284
  id: "google/gemma-4-26b-a4b-it:free",
@@ -10759,9 +11357,9 @@ export const MODELS = {
10759
11357
  reasoning: true,
10760
11358
  input: ["text"],
10761
11359
  cost: {
10762
- input: 0.09999999999999999,
10763
- output: 0.15,
10764
- cacheRead: 0.049999999999999996,
11360
+ input: 0.06,
11361
+ output: 0.25,
11362
+ cacheRead: 0.015,
10765
11363
  cacheWrite: 0,
10766
11364
  },
10767
11365
  contextWindow: 131072,
@@ -10784,9 +11382,9 @@ export const MODELS = {
10784
11382
  contextWindow: 128000,
10785
11383
  maxTokens: 50000,
10786
11384
  },
10787
- "inception/mercury-2.5-preview": {
10788
- id: "inception/mercury-2.5-preview",
10789
- name: "Inception: Mercury 2.5 Preview",
11385
+ "inception/mercury-2.5": {
11386
+ id: "inception/mercury-2.5",
11387
+ name: "Inception: Mercury 2.5",
10790
11388
  api: "openai-completions",
10791
11389
  provider: "openrouter",
10792
11390
  baseUrl: "https://openrouter.ai/api/v1",
@@ -10842,24 +11440,58 @@ export const MODELS = {
10842
11440
  provider: "openrouter",
10843
11441
  baseUrl: "https://openrouter.ai/api/v1",
10844
11442
  reasoning: true,
10845
- input: ["text"],
11443
+ input: ["text"],
11444
+ cost: {
11445
+ input: 0,
11446
+ output: 0,
11447
+ cacheRead: 0,
11448
+ cacheWrite: 0,
11449
+ },
11450
+ contextWindow: 262144,
11451
+ maxTokens: 32768,
11452
+ },
11453
+ "inclusionai/ling-3.0-flash-sante:free": {
11454
+ id: "inclusionai/ling-3.0-flash-sante:free",
11455
+ name: "inclusionAI: Ling 3.0 Flash Sante (free)",
11456
+ api: "openai-completions",
11457
+ provider: "openrouter",
11458
+ baseUrl: "https://openrouter.ai/api/v1",
11459
+ reasoning: true,
11460
+ input: ["text"],
11461
+ cost: {
11462
+ input: 0,
11463
+ output: 0,
11464
+ cacheRead: 0,
11465
+ cacheWrite: 0,
11466
+ },
11467
+ contextWindow: 262144,
11468
+ maxTokens: 32768,
11469
+ },
11470
+ "inclusionai/ling-3.0-flash-vl": {
11471
+ id: "inclusionai/ling-3.0-flash-vl",
11472
+ name: "inclusionAI: Ling 3.0 Flash VL",
11473
+ api: "openai-completions",
11474
+ provider: "openrouter",
11475
+ baseUrl: "https://openrouter.ai/api/v1",
11476
+ reasoning: true,
11477
+ input: ["text", "image"],
10846
11478
  cost: {
10847
- input: 0,
10848
- output: 0,
10849
- cacheRead: 0,
11479
+ input: 0.06,
11480
+ output: 0.18,
11481
+ cacheRead: 0.012,
10850
11482
  cacheWrite: 0,
10851
11483
  },
10852
- contextWindow: 262144,
11484
+ contextWindow: 131072,
10853
11485
  maxTokens: 32768,
10854
11486
  },
10855
- "inclusionai/ling-3.0-flash-sante:free": {
10856
- id: "inclusionai/ling-3.0-flash-sante:free",
10857
- name: "inclusionAI: Ling 3.0 Flash Sante (free)",
11487
+ "inclusionai/ling-3.0-flash-vl:free": {
11488
+ id: "inclusionai/ling-3.0-flash-vl:free",
11489
+ name: "inclusionAI: Ling 3.0 Flash VL (free)",
10858
11490
  api: "openai-completions",
10859
11491
  provider: "openrouter",
10860
11492
  baseUrl: "https://openrouter.ai/api/v1",
10861
11493
  reasoning: true,
10862
- input: ["text"],
11494
+ input: ["text", "image"],
10863
11495
  cost: {
10864
11496
  input: 0,
10865
11497
  output: 0,
@@ -11031,8 +11663,8 @@ export const MODELS = {
11031
11663
  reasoning: true,
11032
11664
  input: ["text", "image"],
11033
11665
  cost: {
11034
- input: 0.3,
11035
- output: 1.1,
11666
+ input: 0.35,
11667
+ output: 1.5,
11036
11668
  cacheRead: 0.04,
11037
11669
  cacheWrite: 0,
11038
11670
  },
@@ -11048,9 +11680,9 @@ export const MODELS = {
11048
11680
  reasoning: true,
11049
11681
  input: ["text", "image"],
11050
11682
  cost: {
11051
- input: 0.35,
11052
- output: 1.5,
11053
- cacheRead: 0.04,
11683
+ input: 0.175,
11684
+ output: 0.75,
11685
+ cacheRead: 0.02,
11054
11686
  cacheWrite: 0,
11055
11687
  },
11056
11688
  contextWindow: 131072,
@@ -11226,23 +11858,6 @@ export const MODELS = {
11226
11858
  contextWindow: 204800,
11227
11859
  maxTokens: 131072,
11228
11860
  },
11229
- "minimax/minimax-m2.7:free": {
11230
- id: "minimax/minimax-m2.7:free",
11231
- name: "MiniMax: MiniMax M2.7 (free)",
11232
- api: "openai-completions",
11233
- provider: "openrouter",
11234
- baseUrl: "https://openrouter.ai/api/v1",
11235
- reasoning: true,
11236
- input: ["text"],
11237
- cost: {
11238
- input: 0,
11239
- output: 0,
11240
- cacheRead: 0,
11241
- cacheWrite: 0,
11242
- },
11243
- contextWindow: 196608,
11244
- maxTokens: 176947,
11245
- },
11246
11861
  "minimax/minimax-m3": {
11247
11862
  id: "minimax/minimax-m3",
11248
11863
  name: "MiniMax: MiniMax M3",
@@ -11277,35 +11892,35 @@ export const MODELS = {
11277
11892
  contextWindow: 524288,
11278
11893
  maxTokens: 471859,
11279
11894
  },
11280
- "minimax/minimax-m3:free": {
11281
- id: "minimax/minimax-m3:free",
11282
- name: "MiniMax: MiniMax M3 (free)",
11895
+ "mistralai/codestral-2508": {
11896
+ id: "mistralai/codestral-2508",
11897
+ name: "Mistral: Codestral 2508",
11283
11898
  api: "openai-completions",
11284
11899
  provider: "openrouter",
11285
11900
  baseUrl: "https://openrouter.ai/api/v1",
11286
- reasoning: true,
11287
- input: ["text", "image"],
11901
+ reasoning: false,
11902
+ input: ["text"],
11288
11903
  cost: {
11289
- input: 0,
11290
- output: 0,
11291
- cacheRead: 0,
11904
+ input: 0.3,
11905
+ output: 0.8999999999999999,
11906
+ cacheRead: 0.03,
11292
11907
  cacheWrite: 0,
11293
11908
  },
11294
- contextWindow: 1048576,
11295
- maxTokens: 943718,
11909
+ contextWindow: 256000,
11910
+ maxTokens: 204800,
11296
11911
  },
11297
- "mistralai/codestral-2508": {
11298
- id: "mistralai/codestral-2508",
11299
- name: "Mistral: Codestral 2508",
11912
+ "mistralai/codestral-2508:batch": {
11913
+ id: "mistralai/codestral-2508:batch",
11914
+ name: "Mistral: Codestral 2508 (batch)",
11300
11915
  api: "openai-completions",
11301
11916
  provider: "openrouter",
11302
11917
  baseUrl: "https://openrouter.ai/api/v1",
11303
11918
  reasoning: false,
11304
11919
  input: ["text"],
11305
11920
  cost: {
11306
- input: 0.3,
11307
- output: 0.8999999999999999,
11308
- cacheRead: 0.03,
11921
+ input: 0.15,
11922
+ output: 0.44999999999999996,
11923
+ cacheRead: 0.015,
11309
11924
  cacheWrite: 0,
11310
11925
  },
11311
11926
  contextWindow: 256000,
@@ -11379,6 +11994,23 @@ export const MODELS = {
11379
11994
  contextWindow: 262144,
11380
11995
  maxTokens: 209715,
11381
11996
  },
11997
+ "mistralai/ministral-8b-2512:batch": {
11998
+ id: "mistralai/ministral-8b-2512:batch",
11999
+ name: "Mistral: Ministral 3 8B 2512 (batch)",
12000
+ api: "openai-completions",
12001
+ provider: "openrouter",
12002
+ baseUrl: "https://openrouter.ai/api/v1",
12003
+ reasoning: false,
12004
+ input: ["text", "image"],
12005
+ cost: {
12006
+ input: 0.075,
12007
+ output: 0.075,
12008
+ cacheRead: 0.0075,
12009
+ cacheWrite: 0,
12010
+ },
12011
+ contextWindow: 262144,
12012
+ maxTokens: 209715,
12013
+ },
11382
12014
  "mistralai/mistral-large": {
11383
12015
  id: "mistralai/mistral-large",
11384
12016
  name: "Mistral Large",
@@ -11430,6 +12062,23 @@ export const MODELS = {
11430
12062
  contextWindow: 262144,
11431
12063
  maxTokens: 209715,
11432
12064
  },
12065
+ "mistralai/mistral-large-2512:batch": {
12066
+ id: "mistralai/mistral-large-2512:batch",
12067
+ name: "Mistral: Mistral Large 3 2512 (batch)",
12068
+ api: "openai-completions",
12069
+ provider: "openrouter",
12070
+ baseUrl: "https://openrouter.ai/api/v1",
12071
+ reasoning: false,
12072
+ input: ["text", "image"],
12073
+ cost: {
12074
+ input: 0.25,
12075
+ output: 0.75,
12076
+ cacheRead: 0.024999999999999998,
12077
+ cacheWrite: 0,
12078
+ },
12079
+ contextWindow: 262144,
12080
+ maxTokens: 209715,
12081
+ },
11433
12082
  "mistralai/mistral-medium-3": {
11434
12083
  id: "mistralai/mistral-medium-3",
11435
12084
  name: "Mistral: Mistral Medium 3",
@@ -11478,8 +12127,8 @@ export const MODELS = {
11478
12127
  cacheRead: 0,
11479
12128
  cacheWrite: 0,
11480
12129
  },
11481
- contextWindow: 32768,
11482
- maxTokens: 26214,
12130
+ contextWindow: 262144,
12131
+ maxTokens: 209715,
11483
12132
  },
11484
12133
  "mistralai/mistral-medium-3.1": {
11485
12134
  id: "mistralai/mistral-medium-3.1",
@@ -11498,6 +12147,23 @@ export const MODELS = {
11498
12147
  contextWindow: 131072,
11499
12148
  maxTokens: 104857,
11500
12149
  },
12150
+ "mistralai/mistral-medium-3.1:batch": {
12151
+ id: "mistralai/mistral-medium-3.1:batch",
12152
+ name: "Mistral: Mistral Medium 3.1 (batch)",
12153
+ api: "openai-completions",
12154
+ provider: "openrouter",
12155
+ baseUrl: "https://openrouter.ai/api/v1",
12156
+ reasoning: false,
12157
+ input: ["text", "image"],
12158
+ cost: {
12159
+ input: 0.19999999999999998,
12160
+ output: 1,
12161
+ cacheRead: 0.02,
12162
+ cacheWrite: 0,
12163
+ },
12164
+ contextWindow: 131072,
12165
+ maxTokens: 104857,
12166
+ },
11501
12167
  "mistralai/mistral-nemo": {
11502
12168
  id: "mistralai/mistral-nemo",
11503
12169
  name: "Mistral: Mistral Nemo",
@@ -11549,6 +12215,23 @@ export const MODELS = {
11549
12215
  contextWindow: 262144,
11550
12216
  maxTokens: 209715,
11551
12217
  },
12218
+ "mistralai/mistral-small-2603:batch": {
12219
+ id: "mistralai/mistral-small-2603:batch",
12220
+ name: "Mistral: Mistral Small 4 (batch)",
12221
+ api: "openai-completions",
12222
+ provider: "openrouter",
12223
+ baseUrl: "https://openrouter.ai/api/v1",
12224
+ reasoning: true,
12225
+ input: ["text", "image"],
12226
+ cost: {
12227
+ input: 0.075,
12228
+ output: 0.3,
12229
+ cacheRead: 0.0075,
12230
+ cacheWrite: 0,
12231
+ },
12232
+ contextWindow: 262144,
12233
+ maxTokens: 209715,
12234
+ },
11552
12235
  "mistralai/mistral-small-3.2-24b-instruct": {
11553
12236
  id: "mistralai/mistral-small-3.2-24b-instruct",
11554
12237
  name: "Mistral: Mistral Small 3.2 24B",
@@ -11563,7 +12246,7 @@ export const MODELS = {
11563
12246
  cacheRead: 0,
11564
12247
  cacheWrite: 0,
11565
12248
  },
11566
- contextWindow: 131072,
12249
+ contextWindow: 256000,
11567
12250
  maxTokens: 16384,
11568
12251
  },
11569
12252
  "mistralai/mixtral-8x22b-instruct": {
@@ -11615,7 +12298,7 @@ export const MODELS = {
11615
12298
  cacheWrite: 0,
11616
12299
  },
11617
12300
  contextWindow: 131072,
11618
- maxTokens: 100352,
12301
+ maxTokens: 98304,
11619
12302
  },
11620
12303
  "moonshotai/kimi-k2-0905": {
11621
12304
  id: "moonshotai/kimi-k2-0905",
@@ -11632,7 +12315,7 @@ export const MODELS = {
11632
12315
  cacheWrite: 0,
11633
12316
  },
11634
12317
  contextWindow: 262144,
11635
- maxTokens: 100352,
12318
+ maxTokens: 98304,
11636
12319
  },
11637
12320
  "moonshotai/kimi-k2-thinking": {
11638
12321
  id: "moonshotai/kimi-k2-thinking",
@@ -11649,7 +12332,7 @@ export const MODELS = {
11649
12332
  cacheWrite: 0,
11650
12333
  },
11651
12334
  contextWindow: 262144,
11652
- maxTokens: 100352,
12335
+ maxTokens: 98304,
11653
12336
  },
11654
12337
  "moonshotai/kimi-k2.5": {
11655
12338
  id: "moonshotai/kimi-k2.5",
@@ -11694,9 +12377,9 @@ export const MODELS = {
11694
12377
  reasoning: true,
11695
12378
  input: ["text", "image"],
11696
12379
  cost: {
11697
- input: 0.66,
11698
- output: 3.4,
11699
- cacheRead: 0.18,
12380
+ input: 0.71,
12381
+ output: 3.5,
12382
+ cacheRead: 0.15,
11700
12383
  cacheWrite: 0,
11701
12384
  },
11702
12385
  contextWindow: 262144,
@@ -11711,9 +12394,9 @@ export const MODELS = {
11711
12394
  reasoning: true,
11712
12395
  input: ["text", "image"],
11713
12396
  cost: {
11714
- input: 3,
11715
- output: 15,
11716
- cacheRead: 0.3,
12397
+ input: 2.6481380629999998,
12398
+ output: 13.282724250000001,
12399
+ cacheRead: 0.30264435,
11717
12400
  cacheWrite: 0,
11718
12401
  },
11719
12402
  contextWindow: 1048576,
@@ -11736,35 +12419,35 @@ export const MODELS = {
11736
12419
  contextWindow: 1048576,
11737
12420
  maxTokens: 943718,
11738
12421
  },
11739
- "nex-agi/nex-n2-mini": {
11740
- id: "nex-agi/nex-n2-mini",
11741
- name: "Nex AGI: Nex-N2-Mini",
12422
+ "nex-agi/nex-n2.5-mini:free": {
12423
+ id: "nex-agi/nex-n2.5-mini:free",
12424
+ name: "Nex AGI: Nex-N2.5-Mini (free)",
11742
12425
  api: "openai-completions",
11743
12426
  provider: "openrouter",
11744
12427
  baseUrl: "https://openrouter.ai/api/v1",
11745
12428
  reasoning: true,
11746
12429
  input: ["text", "image"],
11747
12430
  cost: {
11748
- input: 0.024999999999999998,
11749
- output: 0.09999999999999999,
11750
- cacheRead: 0.0025,
12431
+ input: 0,
12432
+ output: 0,
12433
+ cacheRead: 0,
11751
12434
  cacheWrite: 0,
11752
12435
  },
11753
12436
  contextWindow: 262144,
11754
12437
  maxTokens: 235929,
11755
12438
  },
11756
- "nex-agi/nex-n2-pro": {
11757
- id: "nex-agi/nex-n2-pro",
11758
- name: "Nex AGI: Nex-N2-Pro",
12439
+ "nex-agi/nex-n2.5-pro:free": {
12440
+ id: "nex-agi/nex-n2.5-pro:free",
12441
+ name: "Nex AGI: Nex-N2.5-Pro (free)",
11759
12442
  api: "openai-completions",
11760
12443
  provider: "openrouter",
11761
12444
  baseUrl: "https://openrouter.ai/api/v1",
11762
12445
  reasoning: true,
11763
12446
  input: ["text", "image"],
11764
12447
  cost: {
11765
- input: 0.25,
11766
- output: 1,
11767
- cacheRead: 0.024999999999999998,
12448
+ input: 0,
12449
+ output: 0,
12450
+ cacheRead: 0,
11768
12451
  cacheWrite: 0,
11769
12452
  },
11770
12453
  contextWindow: 262144,
@@ -11818,7 +12501,7 @@ export const MODELS = {
11818
12501
  cacheRead: 0,
11819
12502
  cacheWrite: 0,
11820
12503
  },
11821
- contextWindow: 1000000,
12504
+ contextWindow: 262144,
11822
12505
  maxTokens: 16384,
11823
12506
  },
11824
12507
  "nvidia/nemotron-3-super-120b-a12b:free": {
@@ -13581,13 +14264,13 @@ export const MODELS = {
13581
14264
  reasoning: true,
13582
14265
  input: ["text"],
13583
14266
  cost: {
13584
- input: 0.12,
13585
- output: 0.24,
14267
+ input: 0.22749999999999998,
14268
+ output: 0.9099999999999999,
13586
14269
  cacheRead: 0,
13587
14270
  cacheWrite: 0,
13588
14271
  },
13589
14272
  contextWindow: 131072,
13590
- maxTokens: 16384,
14273
+ maxTokens: 8192,
13591
14274
  },
13592
14275
  "qwen/qwen3-235b-a22b": {
13593
14276
  id: "qwen/qwen3-235b-a22b",
@@ -13615,13 +14298,13 @@ export const MODELS = {
13615
14298
  reasoning: false,
13616
14299
  input: ["text"],
13617
14300
  cost: {
13618
- input: 0.09,
13619
- output: 0.55,
13620
- cacheRead: 0,
14301
+ input: 0.0875,
14302
+ output: 0.35,
14303
+ cacheRead: 0.0175,
13621
14304
  cacheWrite: 0,
13622
14305
  },
13623
14306
  contextWindow: 262144,
13624
- maxTokens: 16384,
14307
+ maxTokens: 235929,
13625
14308
  },
13626
14309
  "qwen/qwen3-235b-a22b-thinking-2507": {
13627
14310
  id: "qwen/qwen3-235b-a22b-thinking-2507",
@@ -13853,13 +14536,13 @@ export const MODELS = {
13853
14536
  reasoning: false,
13854
14537
  input: ["text"],
13855
14538
  cost: {
13856
- input: 0.09999999999999999,
14539
+ input: 0.09,
13857
14540
  output: 1.1,
13858
- cacheRead: 0.07,
14541
+ cacheRead: 0,
13859
14542
  cacheWrite: 0,
13860
14543
  },
13861
14544
  contextWindow: 262144,
13862
- maxTokens: 235929,
14545
+ maxTokens: 16384,
13863
14546
  },
13864
14547
  "qwen/qwen3-next-80b-a3b-thinking": {
13865
14548
  id: "qwen/qwen3-next-80b-a3b-thinking",
@@ -13876,7 +14559,7 @@ export const MODELS = {
13876
14559
  cacheWrite: 0,
13877
14560
  },
13878
14561
  contextWindow: 262144,
13879
- maxTokens: 32768,
14562
+ maxTokens: 235929,
13880
14563
  },
13881
14564
  "qwen/qwen3-vl-235b-a22b-instruct": {
13882
14565
  id: "qwen/qwen3-vl-235b-a22b-instruct",
@@ -14006,13 +14689,13 @@ export const MODELS = {
14006
14689
  reasoning: true,
14007
14690
  input: ["text", "image"],
14008
14691
  cost: {
14009
- input: 0.29,
14010
- output: 2.4,
14692
+ input: 0.26,
14693
+ output: 2.08,
14011
14694
  cacheRead: 0,
14012
14695
  cacheWrite: 0,
14013
14696
  },
14014
14697
  contextWindow: 262144,
14015
- maxTokens: 81920,
14698
+ maxTokens: 65536,
14016
14699
  },
14017
14700
  "qwen/qwen3.5-27b": {
14018
14701
  id: "qwen/qwen3.5-27b",
@@ -14301,7 +14984,7 @@ export const MODELS = {
14301
14984
  cacheWrite: 0,
14302
14985
  },
14303
14986
  contextWindow: 1048576,
14304
- maxTokens: 262144,
14987
+ maxTokens: 131072,
14305
14988
  },
14306
14989
  "qwen/qwen3.8-2.4t-a95b:batch": {
14307
14990
  id: "qwen/qwen3.8-2.4t-a95b:batch",
@@ -14329,9 +15012,9 @@ export const MODELS = {
14329
15012
  reasoning: true,
14330
15013
  input: ["text", "image"],
14331
15014
  cost: {
14332
- input: 0.42,
14333
- output: 3,
14334
- cacheRead: 0.08499999999999999,
15015
+ input: 0.21400000000000002,
15016
+ output: 2.5500000000000003,
15017
+ cacheRead: 0.15,
14335
15018
  cacheWrite: 0,
14336
15019
  },
14337
15020
  contextWindow: 1000000,
@@ -14405,6 +15088,23 @@ export const MODELS = {
14405
15088
  contextWindow: 256000,
14406
15089
  maxTokens: 128000,
14407
15090
  },
15091
+ "sakana/fugu-max": {
15092
+ id: "sakana/fugu-max",
15093
+ name: "Sakana: Fugu Max",
15094
+ api: "openai-completions",
15095
+ provider: "openrouter",
15096
+ baseUrl: "https://openrouter.ai/api/v1",
15097
+ reasoning: true,
15098
+ input: ["text", "image"],
15099
+ cost: {
15100
+ input: 2,
15101
+ output: 6,
15102
+ cacheRead: 0.25,
15103
+ cacheWrite: 0,
15104
+ },
15105
+ contextWindow: 1000000,
15106
+ maxTokens: 128000,
15107
+ },
14408
15108
  "sakana/fugu-ultra": {
14409
15109
  id: "sakana/fugu-ultra",
14410
15110
  name: "Sakana: Fugu Ultra",
@@ -14422,6 +15122,23 @@ export const MODELS = {
14422
15122
  contextWindow: 1000000,
14423
15123
  maxTokens: 128000,
14424
15124
  },
15125
+ "sakana/fugu-ultra-v2": {
15126
+ id: "sakana/fugu-ultra-v2",
15127
+ name: "Sakana: Fugu Ultra v2",
15128
+ api: "openai-completions",
15129
+ provider: "openrouter",
15130
+ baseUrl: "https://openrouter.ai/api/v1",
15131
+ reasoning: true,
15132
+ input: ["text", "image"],
15133
+ cost: {
15134
+ input: 5,
15135
+ output: 30,
15136
+ cacheRead: 0.5,
15137
+ cacheWrite: 0,
15138
+ },
15139
+ contextWindow: 1000000,
15140
+ maxTokens: 128000,
15141
+ },
14425
15142
  "sakana/sakana-namazu": {
14426
15143
  id: "sakana/sakana-namazu",
14427
15144
  name: "Sakana: Sakana Namazu",
@@ -14499,9 +15216,9 @@ export const MODELS = {
14499
15216
  reasoning: true,
14500
15217
  input: ["text"],
14501
15218
  cost: {
14502
- input: 0.13199999999999998,
14503
- output: 0.5279999999999999,
14504
- cacheRead: 0.032999999999999995,
15219
+ input: 0.0825,
15220
+ output: 0.33,
15221
+ cacheRead: 0.020625,
14505
15222
  cacheWrite: 0,
14506
15223
  },
14507
15224
  contextWindow: 262144,
@@ -14686,9 +15403,9 @@ export const MODELS = {
14686
15403
  reasoning: true,
14687
15404
  input: ["text"],
14688
15405
  cost: {
14689
- input: 0.03,
14690
- output: 0.12,
14691
- cacheRead: 0.006,
15406
+ input: 0.09,
15407
+ output: 0.36,
15408
+ cacheRead: 0.018,
14692
15409
  cacheWrite: 0,
14693
15410
  },
14694
15411
  contextWindow: 524288,
@@ -14890,13 +15607,13 @@ export const MODELS = {
14890
15607
  reasoning: true,
14891
15608
  input: ["text"],
14892
15609
  cost: {
14893
- input: 0.55,
14894
- output: 2.2,
14895
- cacheRead: 0.11,
15610
+ input: 0.43,
15611
+ output: 1.75,
15612
+ cacheRead: 0.08,
14896
15613
  cacheWrite: 0,
14897
15614
  },
14898
15615
  contextWindow: 204800,
14899
- maxTokens: 131072,
15616
+ maxTokens: 16384,
14900
15617
  },
14901
15618
  "z-ai/glm-4.6v": {
14902
15619
  id: "z-ai/glm-4.6v",
@@ -14941,13 +15658,13 @@ export const MODELS = {
14941
15658
  reasoning: true,
14942
15659
  input: ["text"],
14943
15660
  cost: {
14944
- input: 0.06,
15661
+ input: 0.060500000000000005,
14945
15662
  output: 0.39999999999999997,
14946
- cacheRead: 0.01,
15663
+ cacheRead: 0,
14947
15664
  cacheWrite: 0,
14948
15665
  },
14949
- contextWindow: 202752,
14950
- maxTokens: 16384,
15666
+ contextWindow: 200000,
15667
+ maxTokens: 117964,
14951
15668
  },
14952
15669
  "z-ai/glm-5": {
14953
15670
  id: "z-ai/glm-5",
@@ -15009,30 +15726,30 @@ export const MODELS = {
15009
15726
  reasoning: true,
15010
15727
  input: ["text"],
15011
15728
  cost: {
15012
- input: 0.966,
15013
- output: 3.036,
15014
- cacheRead: 0.1932,
15729
+ input: 0.6,
15730
+ output: 2,
15731
+ cacheRead: 0.15,
15015
15732
  cacheWrite: 0,
15016
15733
  },
15017
15734
  contextWindow: 1048576,
15018
- maxTokens: 131072,
15735
+ maxTokens: 182476,
15019
15736
  },
15020
- "z-ai/glm-5.2:free": {
15021
- id: "z-ai/glm-5.2:free",
15022
- name: "Z.ai: GLM 5.2 (free)",
15737
+ "z-ai/glm-5.2:batch": {
15738
+ id: "z-ai/glm-5.2:batch",
15739
+ name: "Z.ai: GLM 5.2 (batch)",
15023
15740
  api: "openai-completions",
15024
15741
  provider: "openrouter",
15025
15742
  baseUrl: "https://openrouter.ai/api/v1",
15026
15743
  reasoning: true,
15027
15744
  input: ["text"],
15028
15745
  cost: {
15029
- input: 0,
15030
- output: 0,
15031
- cacheRead: 0,
15746
+ input: 0.7,
15747
+ output: 2.2,
15748
+ cacheRead: 0.07,
15032
15749
  cacheWrite: 0,
15033
15750
  },
15034
- contextWindow: 256000,
15035
- maxTokens: 230400,
15751
+ contextWindow: 1048576,
15752
+ maxTokens: 943718,
15036
15753
  },
15037
15754
  "z-ai/glm-5.3": {
15038
15755
  id: "z-ai/glm-5.3",
@@ -15043,17 +15760,34 @@ export const MODELS = {
15043
15760
  reasoning: true,
15044
15761
  input: ["text"],
15045
15762
  cost: {
15046
- input: 1.4,
15047
- output: 4.4,
15048
- cacheRead: 0.14,
15763
+ input: 1.092,
15764
+ output: 3.432,
15765
+ cacheRead: 0.20279999999999998,
15766
+ cacheWrite: 0,
15767
+ },
15768
+ contextWindow: 1310720,
15769
+ maxTokens: 131072,
15770
+ },
15771
+ "z-ai/glm-5.3-flash": {
15772
+ id: "z-ai/glm-5.3-flash",
15773
+ name: "Z.ai: GLM 5.3 Flash",
15774
+ api: "openai-completions",
15775
+ provider: "openrouter",
15776
+ baseUrl: "https://openrouter.ai/api/v1",
15777
+ reasoning: true,
15778
+ input: ["text", "image"],
15779
+ cost: {
15780
+ input: 0.15,
15781
+ output: 0.5,
15782
+ cacheRead: 0.03,
15049
15783
  cacheWrite: 0,
15050
15784
  },
15051
15785
  contextWindow: 1310720,
15052
- maxTokens: 262144,
15786
+ maxTokens: 131072,
15053
15787
  },
15054
- "z-ai/glm-5.3-flash": {
15055
- id: "z-ai/glm-5.3-flash",
15056
- name: "Z.ai: GLM 5.3 Flash",
15788
+ "z-ai/glm-5.3-flash:batch": {
15789
+ id: "z-ai/glm-5.3-flash:batch",
15790
+ name: "Z.ai: GLM 5.3 Flash (batch)",
15057
15791
  api: "openai-completions",
15058
15792
  provider: "openrouter",
15059
15793
  baseUrl: "https://openrouter.ai/api/v1",
@@ -15065,25 +15799,25 @@ export const MODELS = {
15065
15799
  cacheRead: 0.015,
15066
15800
  cacheWrite: 0,
15067
15801
  },
15068
- contextWindow: 1310720,
15069
- maxTokens: 131072,
15802
+ contextWindow: 1048576,
15803
+ maxTokens: 943718,
15070
15804
  },
15071
- "z-ai/glm-5.3-flash:batch": {
15072
- id: "z-ai/glm-5.3-flash:batch",
15073
- name: "Z.ai: GLM 5.3 Flash (batch)",
15805
+ "z-ai/glm-5.3:batch": {
15806
+ id: "z-ai/glm-5.3:batch",
15807
+ name: "Z.ai: GLM 5.3 (batch)",
15074
15808
  api: "openai-completions",
15075
15809
  provider: "openrouter",
15076
15810
  baseUrl: "https://openrouter.ai/api/v1",
15077
15811
  reasoning: true,
15078
- input: ["text", "image"],
15812
+ input: ["text"],
15079
15813
  cost: {
15080
- input: 0.15,
15081
- output: 0.5,
15082
- cacheRead: 0.03,
15814
+ input: 0.7,
15815
+ output: 2.2,
15816
+ cacheRead: 0.13,
15083
15817
  cacheWrite: 0,
15084
15818
  },
15085
- contextWindow: 1048575,
15086
- maxTokens: 943717,
15819
+ contextWindow: 1048576,
15820
+ maxTokens: 943718,
15087
15821
  },
15088
15822
  "z-ai/glm-5v-turbo": {
15089
15823
  id: "z-ai/glm-5v-turbo",
@@ -15121,7 +15855,7 @@ export const MODELS = {
15121
15855
  },
15122
15856
  "~anthropic/claude-haiku-latest": {
15123
15857
  id: "~anthropic/claude-haiku-latest",
15124
- name: "Anthropic Claude Haiku Latest",
15858
+ name: "Anthropic: Claude Haiku Latest",
15125
15859
  api: "openai-completions",
15126
15860
  provider: "openrouter",
15127
15861
  baseUrl: "https://openrouter.ai/api/v1",
@@ -15155,7 +15889,7 @@ export const MODELS = {
15155
15889
  },
15156
15890
  "~anthropic/claude-sonnet-latest": {
15157
15891
  id: "~anthropic/claude-sonnet-latest",
15158
- name: "Anthropic Claude Sonnet Latest",
15892
+ name: "Anthropic: Claude Sonnet Latest",
15159
15893
  api: "openai-completions",
15160
15894
  provider: "openrouter",
15161
15895
  baseUrl: "https://openrouter.ai/api/v1",
@@ -15172,24 +15906,24 @@ export const MODELS = {
15172
15906
  },
15173
15907
  "~deepseek/deepseek-v4-flash-latest": {
15174
15908
  id: "~deepseek/deepseek-v4-flash-latest",
15175
- name: "DeepSeek V4 Flash Latest",
15909
+ name: "DeepSeek: DeepSeek V4 Flash Latest",
15176
15910
  api: "openai-completions",
15177
15911
  provider: "openrouter",
15178
15912
  baseUrl: "https://openrouter.ai/api/v1",
15179
15913
  reasoning: true,
15180
15914
  input: ["text"],
15181
15915
  cost: {
15182
- input: 0.045,
15183
- output: 0.09,
15184
- cacheRead: 0.009,
15916
+ input: 0.035199999999999995,
15917
+ output: 0.1056,
15918
+ cacheRead: 0.0011200000000000001,
15185
15919
  cacheWrite: 0,
15186
15920
  },
15187
15921
  contextWindow: 1310720,
15188
- maxTokens: 943718,
15922
+ maxTokens: 131072,
15189
15923
  },
15190
15924
  "~google/gemini-flash-latest": {
15191
15925
  id: "~google/gemini-flash-latest",
15192
- name: "Google Gemini Flash Latest",
15926
+ name: "Google: Gemini Flash Latest",
15193
15927
  api: "openai-completions",
15194
15928
  provider: "openrouter",
15195
15929
  baseUrl: "https://openrouter.ai/api/v1",
@@ -15206,7 +15940,7 @@ export const MODELS = {
15206
15940
  },
15207
15941
  "~google/gemini-pro-latest": {
15208
15942
  id: "~google/gemini-pro-latest",
15209
- name: "Google Gemini Pro Latest",
15943
+ name: "Google: Gemini Pro Latest",
15210
15944
  api: "openai-completions",
15211
15945
  provider: "openrouter",
15212
15946
  baseUrl: "https://openrouter.ai/api/v1",
@@ -15223,41 +15957,58 @@ export const MODELS = {
15223
15957
  },
15224
15958
  "~moonshotai/kimi-latest": {
15225
15959
  id: "~moonshotai/kimi-latest",
15226
- name: "MoonshotAI Kimi Latest",
15960
+ name: "MoonshotAI: Kimi Latest",
15227
15961
  api: "openai-completions",
15228
15962
  provider: "openrouter",
15229
15963
  baseUrl: "https://openrouter.ai/api/v1",
15230
15964
  reasoning: true,
15231
15965
  input: ["text", "image"],
15232
15966
  cost: {
15233
- input: 2.5500000000000003,
15234
- output: 12.75,
15235
- cacheRead: 0.25599998999999996,
15967
+ input: 2.0999999999999996,
15968
+ output: 10.950000000000001,
15969
+ cacheRead: 0.22999999999999998,
15236
15970
  cacheWrite: 0,
15237
15971
  },
15238
15972
  contextWindow: 1048576,
15239
15973
  maxTokens: 943718,
15240
15974
  },
15241
- "~openai/gpt-latest": {
15242
- id: "~openai/gpt-latest",
15243
- name: "OpenAI GPT Latest",
15975
+ "~openai/gpt-astra-latest": {
15976
+ id: "~openai/gpt-astra-latest",
15977
+ name: "OpenAI: GPT Astra Latest",
15244
15978
  api: "openai-completions",
15245
15979
  provider: "openrouter",
15246
15980
  baseUrl: "https://openrouter.ai/api/v1",
15247
15981
  reasoning: true,
15248
15982
  input: ["text", "image"],
15249
15983
  cost: {
15250
- input: 2,
15251
- output: 10,
15252
- cacheRead: 0.19999999999999998,
15253
- cacheWrite: 2.5,
15984
+ input: 10,
15985
+ output: 50,
15986
+ cacheRead: 1,
15987
+ cacheWrite: 12.5,
15988
+ },
15989
+ contextWindow: 1050000,
15990
+ maxTokens: 128000,
15991
+ },
15992
+ "~openai/gpt-luna-latest": {
15993
+ id: "~openai/gpt-luna-latest",
15994
+ name: "OpenAI: GPT Luna Latest",
15995
+ api: "openai-completions",
15996
+ provider: "openrouter",
15997
+ baseUrl: "https://openrouter.ai/api/v1",
15998
+ reasoning: true,
15999
+ input: ["text", "image"],
16000
+ cost: {
16001
+ input: 0.19999999999999998,
16002
+ output: 1.2,
16003
+ cacheRead: 0.02,
16004
+ cacheWrite: 0.25,
15254
16005
  },
15255
16006
  contextWindow: 1050000,
15256
16007
  maxTokens: 128000,
15257
16008
  },
15258
16009
  "~openai/gpt-mini-latest": {
15259
16010
  id: "~openai/gpt-mini-latest",
15260
- name: "OpenAI GPT Mini Latest",
16011
+ name: "OpenAI: GPT Mini Latest",
15261
16012
  api: "openai-completions",
15262
16013
  provider: "openrouter",
15263
16014
  baseUrl: "https://openrouter.ai/api/v1",
@@ -15272,6 +16023,40 @@ export const MODELS = {
15272
16023
  contextWindow: 400000,
15273
16024
  maxTokens: 128000,
15274
16025
  },
16026
+ "~openai/gpt-sol-latest": {
16027
+ id: "~openai/gpt-sol-latest",
16028
+ name: "OpenAI: GPT Sol Latest",
16029
+ api: "openai-completions",
16030
+ provider: "openrouter",
16031
+ baseUrl: "https://openrouter.ai/api/v1",
16032
+ reasoning: true,
16033
+ input: ["text", "image"],
16034
+ cost: {
16035
+ input: 2,
16036
+ output: 10,
16037
+ cacheRead: 0.19999999999999998,
16038
+ cacheWrite: 2.5,
16039
+ },
16040
+ contextWindow: 1050000,
16041
+ maxTokens: 128000,
16042
+ },
16043
+ "~openai/gpt-terra-latest": {
16044
+ id: "~openai/gpt-terra-latest",
16045
+ name: "OpenAI: GPT Terra Latest",
16046
+ api: "openai-completions",
16047
+ provider: "openrouter",
16048
+ baseUrl: "https://openrouter.ai/api/v1",
16049
+ reasoning: true,
16050
+ input: ["text", "image"],
16051
+ cost: {
16052
+ input: 2,
16053
+ output: 12,
16054
+ cacheRead: 0.19999999999999998,
16055
+ cacheWrite: 2.5,
16056
+ },
16057
+ contextWindow: 1050000,
16058
+ maxTokens: 128000,
16059
+ },
15275
16060
  "~x-ai/grok-latest": {
15276
16061
  id: "~x-ai/grok-latest",
15277
16062
  name: "xAI: Grok Latest",
@@ -15304,7 +16089,7 @@ export const MODELS = {
15304
16089
  cacheWrite: 0,
15305
16090
  },
15306
16091
  contextWindow: 1310720,
15307
- maxTokens: 943718,
16092
+ maxTokens: 131072,
15308
16093
  },
15309
16094
  "~z-ai/glm-latest": {
15310
16095
  id: "~z-ai/glm-latest",
@@ -15315,9 +16100,9 @@ export const MODELS = {
15315
16100
  reasoning: true,
15316
16101
  input: ["text"],
15317
16102
  cost: {
15318
- input: 1.17,
15319
- output: 3.9600000000000004,
15320
- cacheRead: 0.234,
16103
+ input: 0.936,
16104
+ output: 3.168,
16105
+ cacheRead: 0.1872,
15321
16106
  cacheWrite: 0,
15322
16107
  },
15323
16108
  contextWindow: 1310720,
@@ -15801,23 +16586,6 @@ export const MODELS = {
15801
16586
  contextWindow: 991000,
15802
16587
  maxTokens: 128000,
15803
16588
  },
15804
- "alibaba/qwen3.8-flash-next": {
15805
- id: "alibaba/qwen3.8-flash-next",
15806
- name: "Qwen 3.8 Flash Next",
15807
- api: "anthropic-messages",
15808
- provider: "vercel-ai-gateway",
15809
- baseUrl: "https://ai-gateway.vercel.sh",
15810
- reasoning: true,
15811
- input: ["text", "image"],
15812
- cost: {
15813
- input: 0.12,
15814
- output: 0.39999999999999997,
15815
- cacheRead: 0.01,
15816
- cacheWrite: 0,
15817
- },
15818
- contextWindow: 1048576,
15819
- maxTokens: 1048576,
15820
- },
15821
16589
  "alibaba/qwen3.8-max": {
15822
16590
  id: "alibaba/qwen3.8-max",
15823
16591
  name: "Qwen 3.8 Max",
@@ -16243,6 +17011,23 @@ export const MODELS = {
16243
17011
  contextWindow: 256000,
16244
17012
  maxTokens: 64000,
16245
17013
  },
17014
+ "bytedance/seed-2.1-turbo": {
17015
+ id: "bytedance/seed-2.1-turbo",
17016
+ name: "Seed 2.1 Turbo",
17017
+ api: "anthropic-messages",
17018
+ provider: "vercel-ai-gateway",
17019
+ baseUrl: "https://ai-gateway.vercel.sh",
17020
+ reasoning: true,
17021
+ input: ["text", "image"],
17022
+ cost: {
17023
+ input: 0.5,
17024
+ output: 2.5,
17025
+ cacheRead: 0.09999999999999999,
17026
+ cacheWrite: 0,
17027
+ },
17028
+ contextWindow: 262144,
17029
+ maxTokens: 262144,
17030
+ },
16246
17031
  "cohere/command-a": {
16247
17032
  id: "cohere/command-a",
16248
17033
  name: "Command A",
@@ -16320,9 +17105,9 @@ export const MODELS = {
16320
17105
  reasoning: false,
16321
17106
  input: ["text"],
16322
17107
  cost: {
16323
- input: 0.28,
16324
- output: 0.42,
16325
- cacheRead: 0.028,
17108
+ input: 0.62,
17109
+ output: 1.85,
17110
+ cacheRead: 0,
16326
17111
  cacheWrite: 0,
16327
17112
  },
16328
17113
  contextWindow: 128000,
@@ -16430,6 +17215,23 @@ export const MODELS = {
16430
17215
  contextWindow: 1000000,
16431
17216
  maxTokens: 384000,
16432
17217
  },
17218
+ "deepseek/deepseek-v4.1-flash": {
17219
+ id: "deepseek/deepseek-v4.1-flash",
17220
+ name: "DeepSeek V4.1 Flash",
17221
+ api: "anthropic-messages",
17222
+ provider: "vercel-ai-gateway",
17223
+ baseUrl: "https://ai-gateway.vercel.sh",
17224
+ reasoning: true,
17225
+ input: ["text", "image"],
17226
+ cost: {
17227
+ input: 0.15,
17228
+ output: 0.6,
17229
+ cacheRead: 0.015,
17230
+ cacheWrite: 0,
17231
+ },
17232
+ contextWindow: 1000000,
17233
+ maxTokens: 384000,
17234
+ },
16433
17235
  "google/gemini-2.5-flash": {
16434
17236
  id: "google/gemini-2.5-flash",
16435
17237
  name: "Gemini 2.5 Flash",
@@ -16668,6 +17470,23 @@ export const MODELS = {
16668
17470
  contextWindow: 128000,
16669
17471
  maxTokens: 128000,
16670
17472
  },
17473
+ "inception/mercury-2.5": {
17474
+ id: "inception/mercury-2.5",
17475
+ name: "Mercury 2.5",
17476
+ api: "anthropic-messages",
17477
+ provider: "vercel-ai-gateway",
17478
+ baseUrl: "https://ai-gateway.vercel.sh",
17479
+ reasoning: true,
17480
+ input: ["text"],
17481
+ cost: {
17482
+ input: 0.04,
17483
+ output: 0.15,
17484
+ cacheRead: 0.004,
17485
+ cacheWrite: 0,
17486
+ },
17487
+ contextWindow: 260000,
17488
+ maxTokens: 65536,
17489
+ },
16671
17490
  "inception/mercury-coder-small": {
16672
17491
  id: "inception/mercury-coder-small",
16673
17492
  name: "Mercury Coder Small Beta",
@@ -16770,6 +17589,40 @@ export const MODELS = {
16770
17589
  contextWindow: 256000,
16771
17590
  maxTokens: 32000,
16772
17591
  },
17592
+ "inclusionai/ling-3.0-flash-vl": {
17593
+ id: "inclusionai/ling-3.0-flash-vl",
17594
+ name: "Ling 3.0 Flash VL",
17595
+ api: "anthropic-messages",
17596
+ provider: "vercel-ai-gateway",
17597
+ baseUrl: "https://ai-gateway.vercel.sh",
17598
+ reasoning: true,
17599
+ input: ["text", "image"],
17600
+ cost: {
17601
+ input: 0,
17602
+ output: 0,
17603
+ cacheRead: 0,
17604
+ cacheWrite: 0,
17605
+ },
17606
+ contextWindow: 256000,
17607
+ maxTokens: 32000,
17608
+ },
17609
+ "inclusionai/ling-3.0-flash-vl-free": {
17610
+ id: "inclusionai/ling-3.0-flash-vl-free",
17611
+ name: "Ling 3.0 Flash VL (Free)",
17612
+ api: "anthropic-messages",
17613
+ provider: "vercel-ai-gateway",
17614
+ baseUrl: "https://ai-gateway.vercel.sh",
17615
+ reasoning: true,
17616
+ input: ["text", "image"],
17617
+ cost: {
17618
+ input: 0,
17619
+ output: 0,
17620
+ cacheRead: 0,
17621
+ cacheWrite: 0,
17622
+ },
17623
+ contextWindow: 256000,
17624
+ maxTokens: 32000,
17625
+ },
16773
17626
  "interfaze/interfaze-beta": {
16774
17627
  id: "interfaze/interfaze-beta",
16775
17628
  name: "Interfaze Beta",
@@ -17144,23 +17997,6 @@ export const MODELS = {
17144
17997
  contextWindow: 204800,
17145
17998
  maxTokens: 131000,
17146
17999
  },
17147
- "minimax/minimax-m2.7-free": {
17148
- id: "minimax/minimax-m2.7-free",
17149
- name: "MiniMax M2.7 (Free)",
17150
- api: "anthropic-messages",
17151
- provider: "vercel-ai-gateway",
17152
- baseUrl: "https://ai-gateway.vercel.sh",
17153
- reasoning: true,
17154
- input: ["text"],
17155
- cost: {
17156
- input: 0,
17157
- output: 0,
17158
- cacheRead: 0,
17159
- cacheWrite: 0,
17160
- },
17161
- contextWindow: 196608,
17162
- maxTokens: 196608,
17163
- },
17164
18000
  "minimax/minimax-m2.7-highspeed": {
17165
18001
  id: "minimax/minimax-m2.7-highspeed",
17166
18002
  name: "MiniMax M2.7 High Speed",
@@ -17195,23 +18031,6 @@ export const MODELS = {
17195
18031
  contextWindow: 512000,
17196
18032
  maxTokens: 512000,
17197
18033
  },
17198
- "minimax/minimax-m3-free": {
17199
- id: "minimax/minimax-m3-free",
17200
- name: "MiniMax M3 (Free)",
17201
- api: "anthropic-messages",
17202
- provider: "vercel-ai-gateway",
17203
- baseUrl: "https://ai-gateway.vercel.sh",
17204
- reasoning: true,
17205
- input: ["text", "image"],
17206
- cost: {
17207
- input: 0,
17208
- output: 0,
17209
- cacheRead: 0,
17210
- cacheWrite: 0,
17211
- },
17212
- contextWindow: 1048576,
17213
- maxTokens: 1048576,
17214
- },
17215
18034
  "mistral/codestral": {
17216
18035
  id: "mistral/codestral",
17217
18036
  name: "Mistral Codestral",
@@ -18674,6 +19493,23 @@ export const MODELS = {
18674
19493
  contextWindow: 256000,
18675
19494
  maxTokens: 32768,
18676
19495
  },
19496
+ "sakana/fugu-max": {
19497
+ id: "sakana/fugu-max",
19498
+ name: "Fugu Max",
19499
+ api: "anthropic-messages",
19500
+ provider: "vercel-ai-gateway",
19501
+ baseUrl: "https://ai-gateway.vercel.sh",
19502
+ reasoning: true,
19503
+ input: ["text", "image"],
19504
+ cost: {
19505
+ input: 2,
19506
+ output: 6,
19507
+ cacheRead: 0.25,
19508
+ cacheWrite: 0,
19509
+ },
19510
+ contextWindow: 1000000,
19511
+ maxTokens: 1000000,
19512
+ },
18677
19513
  "sakana/fugu-ultra": {
18678
19514
  id: "sakana/fugu-ultra",
18679
19515
  name: "Fugu Ultra",
@@ -18691,6 +19527,23 @@ export const MODELS = {
18691
19527
  contextWindow: 1000000,
18692
19528
  maxTokens: 1000000,
18693
19529
  },
19530
+ "sakana/fugu-ultra-v2": {
19531
+ id: "sakana/fugu-ultra-v2",
19532
+ name: "Fugu Ultra v2",
19533
+ api: "anthropic-messages",
19534
+ provider: "vercel-ai-gateway",
19535
+ baseUrl: "https://ai-gateway.vercel.sh",
19536
+ reasoning: true,
19537
+ input: ["text", "image"],
19538
+ cost: {
19539
+ input: 5,
19540
+ output: 30,
19541
+ cacheRead: 0.5,
19542
+ cacheWrite: 0,
19543
+ },
19544
+ contextWindow: 1000000,
19545
+ maxTokens: 1000000,
19546
+ },
18694
19547
  "sakana/namazu": {
18695
19548
  id: "sakana/namazu",
18696
19549
  name: "Sakana Namazu",
@@ -19048,23 +19901,6 @@ export const MODELS = {
19048
19901
  contextWindow: 1050000,
19049
19902
  maxTokens: 131000,
19050
19903
  },
19051
- "xiaomi/mimo-v2.5-pro-ultraspeed": {
19052
- id: "xiaomi/mimo-v2.5-pro-ultraspeed",
19053
- name: "MiMo V2.5 Pro UltraSpeed",
19054
- api: "anthropic-messages",
19055
- provider: "vercel-ai-gateway",
19056
- baseUrl: "https://ai-gateway.vercel.sh",
19057
- reasoning: true,
19058
- input: ["text"],
19059
- cost: {
19060
- input: 1.305,
19061
- output: 2.61,
19062
- cacheRead: 0.0108,
19063
- cacheWrite: 0,
19064
- },
19065
- contextWindow: 1048576,
19066
- maxTokens: 131072,
19067
- },
19068
19904
  "zai/glm-4.5": {
19069
19905
  id: "zai/glm-4.5",
19070
19906
  name: "GLM 4.5",
@@ -19278,9 +20114,9 @@ export const MODELS = {
19278
20114
  reasoning: true,
19279
20115
  input: ["text"],
19280
20116
  cost: {
19281
- input: 0.7,
19282
- output: 2.2,
19283
- cacheRead: 0.13,
20117
+ input: 1.4,
20118
+ output: 4.4,
20119
+ cacheRead: 0.14,
19284
20120
  cacheWrite: 0,
19285
20121
  },
19286
20122
  contextWindow: 1000000,
@@ -19320,23 +20156,6 @@ export const MODELS = {
19320
20156
  contextWindow: 1000000,
19321
20157
  maxTokens: 131000,
19322
20158
  },
19323
- "zai/glm-5.3-promo-50": {
19324
- id: "zai/glm-5.3-promo-50",
19325
- name: "GLM 5.3 (50% off)",
19326
- api: "anthropic-messages",
19327
- provider: "vercel-ai-gateway",
19328
- baseUrl: "https://ai-gateway.vercel.sh",
19329
- reasoning: true,
19330
- input: ["text"],
19331
- cost: {
19332
- input: 0.7,
19333
- output: 2.2,
19334
- cacheRead: 0.13,
19335
- cacheWrite: 0,
19336
- },
19337
- contextWindow: 1048576,
19338
- maxTokens: 1048576,
19339
- },
19340
20159
  "zai/glm-5v-turbo": {
19341
20160
  id: "zai/glm-5v-turbo",
19342
20161
  name: "GLM 5V Turbo",