@dreb/ai 2.72.0 → 2.74.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -1362,6 +1362,23 @@ export const MODELS = {
1362
1362
  contextWindow: 1050000,
1363
1363
  maxTokens: 128000,
1364
1364
  },
1365
+ "global.openai.gpt-6.1-terra": {
1366
+ id: "global.openai.gpt-6.1-terra",
1367
+ name: "GPT-6.1 Terra (Global)",
1368
+ api: "bedrock-converse-stream",
1369
+ provider: "amazon-bedrock",
1370
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1371
+ reasoning: true,
1372
+ input: ["text", "image"],
1373
+ cost: {
1374
+ input: 1,
1375
+ output: 5,
1376
+ cacheRead: 0.05,
1377
+ cacheWrite: 1.25,
1378
+ },
1379
+ contextWindow: 1050000,
1380
+ maxTokens: 128000,
1381
+ },
1365
1382
  "global.xai.grok-4.6": {
1366
1383
  id: "global.xai.grok-4.6",
1367
1384
  name: "Grok 4.6 (Global)",
@@ -2263,6 +2280,23 @@ export const MODELS = {
2263
2280
  contextWindow: 1050000,
2264
2281
  maxTokens: 128000,
2265
2282
  },
2283
+ "openai.gpt-6.1-terra": {
2284
+ id: "openai.gpt-6.1-terra",
2285
+ name: "GPT-6.1 Terra",
2286
+ api: "bedrock-converse-stream",
2287
+ provider: "amazon-bedrock",
2288
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2289
+ reasoning: true,
2290
+ input: ["text", "image"],
2291
+ cost: {
2292
+ input: 1.1,
2293
+ output: 5.5,
2294
+ cacheRead: 0.055,
2295
+ cacheWrite: 1.375,
2296
+ },
2297
+ contextWindow: 1050000,
2298
+ maxTokens: 128000,
2299
+ },
2266
2300
  "openai.gpt-oss-120b": {
2267
2301
  id: "openai.gpt-oss-120b",
2268
2302
  name: "gpt-oss-120b",
@@ -3113,6 +3147,23 @@ export const MODELS = {
3113
3147
  contextWindow: 1050000,
3114
3148
  maxTokens: 128000,
3115
3149
  },
3150
+ "us.openai.gpt-6.1-terra": {
3151
+ id: "us.openai.gpt-6.1-terra",
3152
+ name: "GPT-6.1 Terra (US)",
3153
+ api: "bedrock-converse-stream",
3154
+ provider: "amazon-bedrock",
3155
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
3156
+ reasoning: true,
3157
+ input: ["text", "image"],
3158
+ cost: {
3159
+ input: 1.1,
3160
+ output: 5.5,
3161
+ cacheRead: 0.055,
3162
+ cacheWrite: 1.375,
3163
+ },
3164
+ contextWindow: 1050000,
3165
+ maxTokens: 128000,
3166
+ },
3116
3167
  "us.writer.palmyra-x4-v1:0": {
3117
3168
  id: "us.writer.palmyra-x4-v1:0",
3118
3169
  name: "Palmyra X4 (US)",
@@ -7527,6 +7578,23 @@ export const MODELS = {
7527
7578
  contextWindow: 256000,
7528
7579
  maxTokens: 4096,
7529
7580
  },
7581
+ "deepseek-v4-1-flash-1017": {
7582
+ id: "deepseek-v4-1-flash-1017",
7583
+ name: "DeepSeek V4.1 Flash 1017",
7584
+ api: "mistral-conversations",
7585
+ provider: "mistral",
7586
+ baseUrl: "https://api.mistral.ai",
7587
+ reasoning: true,
7588
+ input: ["text"],
7589
+ cost: {
7590
+ input: 0.3,
7591
+ output: 1.2,
7592
+ cacheRead: 0.007,
7593
+ cacheWrite: 0,
7594
+ },
7595
+ contextWindow: 1048576,
7596
+ maxTokens: 128000,
7597
+ },
7530
7598
  "devstral-2512": {
7531
7599
  id: "devstral-2512",
7532
7600
  name: "Devstral 2",
@@ -7757,9 +7825,26 @@ export const MODELS = {
7757
7825
  reasoning: true,
7758
7826
  input: ["text", "image"],
7759
7827
  cost: {
7760
- input: 1.36,
7761
- output: 4.18,
7762
- cacheRead: 0.14,
7828
+ input: 0.7,
7829
+ output: 2.1,
7830
+ cacheRead: 0.07,
7831
+ cacheWrite: 0,
7832
+ },
7833
+ contextWindow: 1048576,
7834
+ maxTokens: 262144,
7835
+ },
7836
+ "mistral-large-4-0-8": {
7837
+ id: "mistral-large-4-0-8",
7838
+ name: "Mistral Large 4.0.8",
7839
+ api: "mistral-conversations",
7840
+ provider: "mistral",
7841
+ baseUrl: "https://api.mistral.ai",
7842
+ reasoning: true,
7843
+ input: ["text", "image"],
7844
+ cost: {
7845
+ input: 0.7,
7846
+ output: 2.1,
7847
+ cacheRead: 0.07,
7763
7848
  cacheWrite: 0,
7764
7849
  },
7765
7850
  contextWindow: 1048576,
@@ -10253,23 +10338,6 @@ export const MODELS = {
10253
10338
  contextWindow: 1048576,
10254
10339
  maxTokens: 131072,
10255
10340
  },
10256
- "nemotron-3-ultra-free": {
10257
- id: "nemotron-3-ultra-free",
10258
- name: "Nemotron 3 Ultra Free",
10259
- api: "openai-completions",
10260
- provider: "opencode",
10261
- baseUrl: "https://opencode.ai/zen/v1",
10262
- reasoning: true,
10263
- input: ["text"],
10264
- cost: {
10265
- input: 0,
10266
- output: 0,
10267
- cacheRead: 0,
10268
- cacheWrite: 0,
10269
- },
10270
- contextWindow: 1000000,
10271
- maxTokens: 128000,
10272
- },
10273
10341
  "nemotron-3.5-lightning-free": {
10274
10342
  id: "nemotron-3.5-lightning-free",
10275
10343
  name: "Nemotron 3.5 Lightning Free",
@@ -11802,7 +11870,7 @@ export const MODELS = {
11802
11870
  cacheWrite: 0,
11803
11871
  },
11804
11872
  contextWindow: 163840,
11805
- maxTokens: 32768,
11873
+ maxTokens: 147456,
11806
11874
  },
11807
11875
  "deepseek/deepseek-r1-0528": {
11808
11876
  id: "deepseek/deepseek-r1-0528",
@@ -11847,13 +11915,13 @@ export const MODELS = {
11847
11915
  reasoning: true,
11848
11916
  input: ["text"],
11849
11917
  cost: {
11850
- input: 0.28,
11851
- output: 0.42,
11852
- cacheRead: 0.028,
11918
+ input: 0.26,
11919
+ output: 0.38,
11920
+ cacheRead: 0.13,
11853
11921
  cacheWrite: 0,
11854
11922
  },
11855
11923
  contextWindow: 163840,
11856
- maxTokens: 65536,
11924
+ maxTokens: 147456,
11857
11925
  },
11858
11926
  "deepseek/deepseek-v3.2-exp": {
11859
11927
  id: "deepseek/deepseek-v3.2-exp",
@@ -11881,13 +11949,13 @@ export const MODELS = {
11881
11949
  reasoning: true,
11882
11950
  input: ["text"],
11883
11951
  cost: {
11884
- input: 0.04732,
11885
- output: 0.09464,
11886
- cacheRead: 0.009464,
11952
+ input: 0.091,
11953
+ output: 0.182,
11954
+ cacheRead: 0.0182,
11887
11955
  cacheWrite: 0,
11888
11956
  },
11889
11957
  contextWindow: 1048576,
11890
- maxTokens: 131072,
11958
+ maxTokens: 943717,
11891
11959
  },
11892
11960
  "deepseek/deepseek-v4-flash-0731": {
11893
11961
  id: "deepseek/deepseek-v4-flash-0731",
@@ -11932,13 +12000,13 @@ export const MODELS = {
11932
12000
  reasoning: true,
11933
12001
  input: ["text"],
11934
12002
  cost: {
11935
- input: 0.88044,
11936
- output: 1.76088,
11937
- cacheRead: 0.07337,
12003
+ input: 1.5999999999999999,
12004
+ output: 3.1999999999999997,
12005
+ cacheRead: 0.135,
11938
12006
  cacheWrite: 0,
11939
12007
  },
11940
12008
  contextWindow: 1048576,
11941
- maxTokens: 384000,
12009
+ maxTokens: 393216,
11942
12010
  },
11943
12011
  "deepseek/deepseek-v4-pro-0813": {
11944
12012
  id: "deepseek/deepseek-v4-pro-0813",
@@ -11949,13 +12017,13 @@ export const MODELS = {
11949
12017
  reasoning: true,
11950
12018
  input: ["text"],
11951
12019
  cost: {
11952
- input: 0.66,
11953
- output: 1.9800000000000002,
11954
- cacheRead: 0.022,
12020
+ input: 0.11499999999999999,
12021
+ output: 4,
12022
+ cacheRead: 0.11,
11955
12023
  cacheWrite: 0,
11956
12024
  },
11957
12025
  contextWindow: 1048576,
11958
- maxTokens: 393216,
12026
+ maxTokens: 943718,
11959
12027
  },
11960
12028
  "deepseek/deepseek-v4.1-flash": {
11961
12029
  id: "deepseek/deepseek-v4.1-flash",
@@ -11966,13 +12034,13 @@ export const MODELS = {
11966
12034
  reasoning: true,
11967
12035
  input: ["text", "image"],
11968
12036
  cost: {
11969
- input: 0.14,
11970
- output: 0.42,
11971
- cacheRead: 0.004200000000000001,
12037
+ input: 0.045,
12038
+ output: 0.5,
12039
+ cacheRead: 0.006,
11972
12040
  cacheWrite: 0,
11973
12041
  },
11974
12042
  contextWindow: 1048576,
11975
- maxTokens: 131072,
12043
+ maxTokens: 943718,
11976
12044
  },
11977
12045
  "deepseek/deepseek-v4.1-flash-1017": {
11978
12046
  id: "deepseek/deepseek-v4.1-flash-1017",
@@ -12613,7 +12681,7 @@ export const MODELS = {
12613
12681
  input: ["text"],
12614
12682
  cost: {
12615
12683
  input: 0.06,
12616
- output: 0.25,
12684
+ output: 0.15,
12617
12685
  cacheRead: 0.015,
12618
12686
  cacheWrite: 0,
12619
12687
  },
@@ -12719,8 +12787,8 @@ export const MODELS = {
12719
12787
  cacheRead: 0.015,
12720
12788
  cacheWrite: 0,
12721
12789
  },
12722
- contextWindow: 262144,
12723
- maxTokens: 235929,
12790
+ contextWindow: 1000000,
12791
+ maxTokens: 262144,
12724
12792
  },
12725
12793
  "liquid/lfm-2.5-2.6b": {
12726
12794
  id: "liquid/lfm-2.5-2.6b",
@@ -13071,13 +13139,13 @@ export const MODELS = {
13071
13139
  reasoning: true,
13072
13140
  input: ["text"],
13073
13141
  cost: {
13074
- input: 0.3,
13075
- output: 1.2,
13142
+ input: 0.27,
13143
+ output: 0.95,
13076
13144
  cacheRead: 0.03,
13077
13145
  cacheWrite: 0,
13078
13146
  },
13079
13147
  contextWindow: 204800,
13080
- maxTokens: 131072,
13148
+ maxTokens: 32768,
13081
13149
  },
13082
13150
  "minimax/minimax-m2.7": {
13083
13151
  id: "minimax/minimax-m2.7",
@@ -13122,9 +13190,9 @@ export const MODELS = {
13122
13190
  reasoning: true,
13123
13191
  input: ["text", "image"],
13124
13192
  cost: {
13125
- input: 0.15,
13126
- output: 0.6,
13127
- cacheRead: 0.015,
13193
+ input: 0.075,
13194
+ output: 0.3,
13195
+ cacheRead: 0.0075,
13128
13196
  cacheWrite: 0,
13129
13197
  },
13130
13198
  contextWindow: 1048576,
@@ -13275,9 +13343,9 @@ export const MODELS = {
13275
13343
  reasoning: true,
13276
13344
  input: ["text", "image"],
13277
13345
  cost: {
13278
- input: 1.3599999999999999,
13279
- output: 4.18,
13280
- cacheRead: 0.14,
13346
+ input: 0.7,
13347
+ output: 2.0999999999999996,
13348
+ cacheRead: 0.07,
13281
13349
  cacheWrite: 0,
13282
13350
  },
13283
13351
  contextWindow: 1048576,
@@ -13564,8 +13632,8 @@ export const MODELS = {
13564
13632
  reasoning: true,
13565
13633
  input: ["text", "image"],
13566
13634
  cost: {
13567
- input: 0.56,
13568
- output: 3.1,
13635
+ input: 0.6,
13636
+ output: 3.5,
13569
13637
  cacheRead: 0.15,
13570
13638
  cacheWrite: 0,
13571
13639
  },
@@ -13581,9 +13649,9 @@ export const MODELS = {
13581
13649
  reasoning: true,
13582
13650
  input: ["text", "image"],
13583
13651
  cost: {
13584
- input: 0.16899999999999998,
13585
- output: 20,
13586
- cacheRead: 0.159,
13652
+ input: 0.192,
13653
+ output: 15,
13654
+ cacheRead: 0.182,
13587
13655
  cacheWrite: 0,
13588
13656
  },
13589
13657
  contextWindow: 1048576,
@@ -13688,8 +13756,8 @@ export const MODELS = {
13688
13756
  cacheRead: 0.15,
13689
13757
  cacheWrite: 0,
13690
13758
  },
13691
- contextWindow: 262144,
13692
- maxTokens: 235929,
13759
+ contextWindow: 1048576,
13760
+ maxTokens: 262144,
13693
13761
  },
13694
13762
  "nvidia/nemotron-3-ultra-550b-a55b": {
13695
13763
  id: "nvidia/nemotron-3-ultra-550b-a55b",
@@ -13844,23 +13912,6 @@ export const MODELS = {
13844
13912
  contextWindow: 8191,
13845
13913
  maxTokens: 4096,
13846
13914
  },
13847
- "openai/gpt-4-turbo": {
13848
- id: "openai/gpt-4-turbo",
13849
- name: "OpenAI: GPT-4 Turbo",
13850
- api: "openai-completions",
13851
- provider: "openrouter",
13852
- baseUrl: "https://openrouter.ai/api/v1",
13853
- reasoning: false,
13854
- input: ["text", "image"],
13855
- cost: {
13856
- input: 10,
13857
- output: 30,
13858
- cacheRead: 0,
13859
- cacheWrite: 0,
13860
- },
13861
- contextWindow: 128000,
13862
- maxTokens: 4096,
13863
- },
13864
13915
  "openai/gpt-4-turbo:batch": {
13865
13916
  id: "openai/gpt-4-turbo:batch",
13866
13917
  name: "OpenAI: GPT-4 Turbo (batch)",
@@ -15298,13 +15349,13 @@ export const MODELS = {
15298
15349
  reasoning: true,
15299
15350
  input: ["text"],
15300
15351
  cost: {
15301
- input: 0.018,
15302
- output: 0.09,
15303
- cacheRead: 0.009,
15352
+ input: 0.03,
15353
+ output: 0.13,
15354
+ cacheRead: 0.03,
15304
15355
  cacheWrite: 0,
15305
15356
  },
15306
15357
  contextWindow: 131072,
15307
- maxTokens: 32768,
15358
+ maxTokens: 117964,
15308
15359
  },
15309
15360
  "openai/gpt-oss-20b:batch": {
15310
15361
  id: "openai/gpt-oss-20b:batch",
@@ -15340,23 +15391,6 @@ export const MODELS = {
15340
15391
  contextWindow: 131072,
15341
15392
  maxTokens: 65536,
15342
15393
  },
15343
- "openai/o1": {
15344
- id: "openai/o1",
15345
- name: "OpenAI: o1",
15346
- api: "openai-completions",
15347
- provider: "openrouter",
15348
- baseUrl: "https://openrouter.ai/api/v1",
15349
- reasoning: true,
15350
- input: ["text", "image"],
15351
- cost: {
15352
- input: 15,
15353
- output: 60,
15354
- cacheRead: 7.5,
15355
- cacheWrite: 0,
15356
- },
15357
- contextWindow: 200000,
15358
- maxTokens: 100000,
15359
- },
15360
15394
  "openai/o3": {
15361
15395
  id: "openai/o3",
15362
15396
  name: "OpenAI: o3",
@@ -15374,40 +15408,6 @@ export const MODELS = {
15374
15408
  contextWindow: 200000,
15375
15409
  maxTokens: 100000,
15376
15410
  },
15377
- "openai/o3-mini": {
15378
- id: "openai/o3-mini",
15379
- name: "OpenAI: o3 Mini",
15380
- api: "openai-completions",
15381
- provider: "openrouter",
15382
- baseUrl: "https://openrouter.ai/api/v1",
15383
- reasoning: true,
15384
- input: ["text"],
15385
- cost: {
15386
- input: 1.1,
15387
- output: 4.4,
15388
- cacheRead: 0.55,
15389
- cacheWrite: 0,
15390
- },
15391
- contextWindow: 200000,
15392
- maxTokens: 100000,
15393
- },
15394
- "openai/o3-mini-high": {
15395
- id: "openai/o3-mini-high",
15396
- name: "OpenAI: o3 Mini High",
15397
- api: "openai-completions",
15398
- provider: "openrouter",
15399
- baseUrl: "https://openrouter.ai/api/v1",
15400
- reasoning: true,
15401
- input: ["text"],
15402
- cost: {
15403
- input: 1.1,
15404
- output: 4.4,
15405
- cacheRead: 0.55,
15406
- cacheWrite: 0,
15407
- },
15408
- contextWindow: 200000,
15409
- maxTokens: 100000,
15410
- },
15411
15411
  "openai/o3-mini:batch": {
15412
15412
  id: "openai/o3-mini:batch",
15413
15413
  name: "OpenAI: o3 Mini (batch)",
@@ -15459,40 +15459,6 @@ export const MODELS = {
15459
15459
  contextWindow: 200000,
15460
15460
  maxTokens: 100000,
15461
15461
  },
15462
- "openai/o4-mini": {
15463
- id: "openai/o4-mini",
15464
- name: "OpenAI: o4 Mini",
15465
- api: "openai-completions",
15466
- provider: "openrouter",
15467
- baseUrl: "https://openrouter.ai/api/v1",
15468
- reasoning: true,
15469
- input: ["text", "image"],
15470
- cost: {
15471
- input: 1.1,
15472
- output: 4.4,
15473
- cacheRead: 0.275,
15474
- cacheWrite: 0,
15475
- },
15476
- contextWindow: 200000,
15477
- maxTokens: 100000,
15478
- },
15479
- "openai/o4-mini-high": {
15480
- id: "openai/o4-mini-high",
15481
- name: "OpenAI: o4 Mini High",
15482
- api: "openai-completions",
15483
- provider: "openrouter",
15484
- baseUrl: "https://openrouter.ai/api/v1",
15485
- reasoning: true,
15486
- input: ["text", "image"],
15487
- cost: {
15488
- input: 1.1,
15489
- output: 4.4,
15490
- cacheRead: 0.275,
15491
- cacheWrite: 0,
15492
- },
15493
- contextWindow: 200000,
15494
- maxTokens: 100000,
15495
- },
15496
15462
  "openai/o4-mini:batch": {
15497
15463
  id: "openai/o4-mini:batch",
15498
15464
  name: "OpenAI: o4 Mini (batch)",
@@ -15542,7 +15508,7 @@ export const MODELS = {
15542
15508
  cacheWrite: 0,
15543
15509
  },
15544
15510
  contextWindow: 2000000,
15545
- maxTokens: 4096,
15511
+ maxTokens: 65536,
15546
15512
  },
15547
15513
  "openrouter/free": {
15548
15514
  id: "openrouter/free",
@@ -15578,6 +15544,23 @@ export const MODELS = {
15578
15544
  contextWindow: 2000000,
15579
15545
  maxTokens: 4096,
15580
15546
  },
15547
+ "particle-ai/particle-1": {
15548
+ id: "particle-ai/particle-1",
15549
+ name: "Particle AI: Particle-1",
15550
+ api: "openai-completions",
15551
+ provider: "openrouter",
15552
+ baseUrl: "https://openrouter.ai/api/v1",
15553
+ reasoning: true,
15554
+ input: ["text", "image"],
15555
+ cost: {
15556
+ input: 0.049999999999999996,
15557
+ output: 0.5,
15558
+ cacheRead: 0.003,
15559
+ cacheWrite: 0,
15560
+ },
15561
+ contextWindow: 1048576,
15562
+ maxTokens: 131072,
15563
+ },
15581
15564
  "perceptron/perceptron-mk1.5": {
15582
15565
  id: "perceptron/perceptron-mk1.5",
15583
15566
  name: "Perceptron: Perceptron Mk1.5",
@@ -15595,90 +15578,39 @@ export const MODELS = {
15595
15578
  contextWindow: 36864,
15596
15579
  maxTokens: 8192,
15597
15580
  },
15598
- "poolside/laguna-s-2.1": {
15599
- id: "poolside/laguna-s-2.1",
15600
- name: "Poolside: Laguna S 2.1",
15601
- api: "openai-completions",
15602
- provider: "openrouter",
15603
- baseUrl: "https://openrouter.ai/api/v1",
15604
- reasoning: true,
15605
- input: ["text"],
15606
- cost: {
15607
- input: 0.09,
15608
- output: 0.18,
15609
- cacheRead: 0.009,
15610
- cacheWrite: 0,
15611
- },
15612
- contextWindow: 1048576,
15613
- maxTokens: 131072,
15614
- },
15615
- "poolside/laguna-s-2.1:free": {
15616
- id: "poolside/laguna-s-2.1:free",
15617
- name: "Poolside: Laguna S 2.1 (free)",
15618
- api: "openai-completions",
15619
- provider: "openrouter",
15620
- baseUrl: "https://openrouter.ai/api/v1",
15621
- reasoning: true,
15622
- input: ["text"],
15623
- cost: {
15624
- input: 0,
15625
- output: 0,
15626
- cacheRead: 0,
15627
- cacheWrite: 0,
15628
- },
15629
- contextWindow: 262144,
15630
- maxTokens: 32768,
15631
- },
15632
- "poolside/laguna-xs-2.1": {
15633
- id: "poolside/laguna-xs-2.1",
15634
- name: "Poolside: Laguna XS 2.1",
15635
- api: "openai-completions",
15636
- provider: "openrouter",
15637
- baseUrl: "https://openrouter.ai/api/v1",
15638
- reasoning: true,
15639
- input: ["text"],
15640
- cost: {
15641
- input: 0.06,
15642
- output: 0.12,
15643
- cacheRead: 0.03,
15644
- cacheWrite: 0,
15645
- },
15646
- contextWindow: 262144,
15647
- maxTokens: 32768,
15648
- },
15649
- "poolside/laguna-xs-2.1:free": {
15650
- id: "poolside/laguna-xs-2.1:free",
15651
- name: "Poolside: Laguna XS 2.1 (free)",
15581
+ "prism-ml/ternary-bonsai-2-27b": {
15582
+ id: "prism-ml/ternary-bonsai-2-27b",
15583
+ name: "PrismML: Ternary Bonsai 2 27B",
15652
15584
  api: "openai-completions",
15653
15585
  provider: "openrouter",
15654
15586
  baseUrl: "https://openrouter.ai/api/v1",
15655
15587
  reasoning: true,
15656
- input: ["text"],
15588
+ input: ["text", "image"],
15657
15589
  cost: {
15658
- input: 0,
15659
- output: 0,
15660
- cacheRead: 0,
15590
+ input: 0.075,
15591
+ output: 0.5,
15592
+ cacheRead: 0.0375,
15661
15593
  cacheWrite: 0,
15662
15594
  },
15663
15595
  contextWindow: 262144,
15664
15596
  maxTokens: 32768,
15665
15597
  },
15666
- "prism-ml/ternary-bonsai-2-27b": {
15667
- id: "prism-ml/ternary-bonsai-2-27b",
15668
- name: "PrismML: Ternary Bonsai 2 27B",
15598
+ "prism-ml/ternary-bonsai-2-9b": {
15599
+ id: "prism-ml/ternary-bonsai-2-9b",
15600
+ name: "PrismML: Ternary Bonsai 2 9B",
15669
15601
  api: "openai-completions",
15670
15602
  provider: "openrouter",
15671
15603
  baseUrl: "https://openrouter.ai/api/v1",
15672
15604
  reasoning: true,
15673
15605
  input: ["text", "image"],
15674
15606
  cost: {
15675
- input: 0.075,
15676
- output: 0.5,
15677
- cacheRead: 0.0375,
15607
+ input: 0.08,
15608
+ output: 0.13,
15609
+ cacheRead: 0.04,
15678
15610
  cacheWrite: 0,
15679
15611
  },
15680
15612
  contextWindow: 262144,
15681
- maxTokens: 32768,
15613
+ maxTokens: 65536,
15682
15614
  },
15683
15615
  "qwen/qwen-2.5-72b-instruct": {
15684
15616
  id: "qwen/qwen-2.5-72b-instruct",
@@ -15918,23 +15850,6 @@ export const MODELS = {
15918
15850
  contextWindow: 262144,
15919
15851
  maxTokens: 16384,
15920
15852
  },
15921
- "qwen/qwen3-next-80b-a3b-thinking": {
15922
- id: "qwen/qwen3-next-80b-a3b-thinking",
15923
- name: "Qwen: Qwen3 Next 80B A3B Thinking",
15924
- api: "openai-completions",
15925
- provider: "openrouter",
15926
- baseUrl: "https://openrouter.ai/api/v1",
15927
- reasoning: true,
15928
- input: ["text"],
15929
- cost: {
15930
- input: 0.15,
15931
- output: 1.2,
15932
- cacheRead: 0,
15933
- cacheWrite: 0,
15934
- },
15935
- contextWindow: 262144,
15936
- maxTokens: 235929,
15937
- },
15938
15853
  "qwen/qwen3-vl-235b-a22b-instruct": {
15939
15854
  id: "qwen/qwen3-vl-235b-a22b-instruct",
15940
15855
  name: "Qwen: Qwen3 VL 235B A22B Instruct",
@@ -16148,13 +16063,13 @@ export const MODELS = {
16148
16063
  reasoning: true,
16149
16064
  input: ["text", "image"],
16150
16065
  cost: {
16151
- input: 0.32,
16152
- output: 3.1999999999999997,
16066
+ input: 0.44999999999999996,
16067
+ output: 2.7,
16153
16068
  cacheRead: 0,
16154
16069
  cacheWrite: 0,
16155
16070
  },
16156
16071
  contextWindow: 262144,
16157
- maxTokens: 81920,
16072
+ maxTokens: 65536,
16158
16073
  },
16159
16074
  "qwen/qwen3.6-35b-a3b": {
16160
16075
  id: "qwen/qwen3.6-35b-a3b",
@@ -16166,8 +16081,8 @@ export const MODELS = {
16166
16081
  input: ["text", "image"],
16167
16082
  cost: {
16168
16083
  input: 0.09999999999999999,
16169
- output: 0.95,
16170
- cacheRead: 0.049999999999999996,
16084
+ output: 0.8999999999999999,
16085
+ cacheRead: 0.024999999999999998,
16171
16086
  cacheWrite: 0,
16172
16087
  },
16173
16088
  contextWindow: 262144,
@@ -16290,7 +16205,7 @@ export const MODELS = {
16290
16205
  cacheWrite: 0,
16291
16206
  },
16292
16207
  contextWindow: 1048576,
16293
- maxTokens: 909000,
16208
+ maxTokens: 131072,
16294
16209
  },
16295
16210
  "qwen/qwen3.8-27b": {
16296
16211
  id: "qwen/qwen3.8-27b",
@@ -16301,12 +16216,12 @@ export const MODELS = {
16301
16216
  reasoning: true,
16302
16217
  input: ["text", "image"],
16303
16218
  cost: {
16304
- input: 0.19999999999999998,
16305
- output: 2.5,
16306
- cacheRead: 0.09,
16219
+ input: 0.15,
16220
+ output: 1.875,
16221
+ cacheRead: 0.0375,
16307
16222
  cacheWrite: 0,
16308
16223
  },
16309
- contextWindow: 1000000,
16224
+ contextWindow: 1024000,
16310
16225
  maxTokens: 235929,
16311
16226
  },
16312
16227
  "qwen/qwen3.8-flash": {
@@ -16751,23 +16666,6 @@ export const MODELS = {
16751
16666
  contextWindow: 524288,
16752
16667
  maxTokens: 131072,
16753
16668
  },
16754
- "upstage/solar-pro-3": {
16755
- id: "upstage/solar-pro-3",
16756
- name: "Upstage: Solar Pro 3",
16757
- api: "openai-completions",
16758
- provider: "openrouter",
16759
- baseUrl: "https://openrouter.ai/api/v1",
16760
- reasoning: true,
16761
- input: ["text"],
16762
- cost: {
16763
- input: 0.15,
16764
- output: 0.6,
16765
- cacheRead: 0.015,
16766
- cacheWrite: 0,
16767
- },
16768
- contextWindow: 131072,
16769
- maxTokens: 117964,
16770
- },
16771
16669
  "upstage/solar-pro4": {
16772
16670
  id: "upstage/solar-pro4",
16773
16671
  name: "Upstage: Solar Pro 4",
@@ -17066,9 +16964,9 @@ export const MODELS = {
17066
16964
  reasoning: true,
17067
16965
  input: ["text"],
17068
16966
  cost: {
17069
- input: 0.55,
17070
- output: 2.2,
17071
- cacheRead: 0.11,
16967
+ input: 0.5,
16968
+ output: 2,
16969
+ cacheRead: 0.09999999999999999,
17072
16970
  cacheWrite: 0,
17073
16971
  },
17074
16972
  contextWindow: 204800,
@@ -17100,9 +16998,9 @@ export const MODELS = {
17100
16998
  reasoning: true,
17101
16999
  input: ["text"],
17102
17000
  cost: {
17103
- input: 0.54,
17104
- output: 1.9800000000000002,
17105
- cacheRead: 0.099,
17001
+ input: 0.6,
17002
+ output: 2.2,
17003
+ cacheRead: 0.11,
17106
17004
  cacheWrite: 0,
17107
17005
  },
17108
17006
  contextWindow: 204800,
@@ -17168,13 +17066,13 @@ export const MODELS = {
17168
17066
  reasoning: true,
17169
17067
  input: ["text"],
17170
17068
  cost: {
17171
- input: 0.966,
17172
- output: 3.036,
17173
- cacheRead: 0.1794,
17069
+ input: 1.4,
17070
+ output: 4.4,
17071
+ cacheRead: 0.26,
17174
17072
  cacheWrite: 0,
17175
17073
  },
17176
17074
  contextWindow: 204800,
17177
- maxTokens: 128000,
17075
+ maxTokens: 131072,
17178
17076
  },
17179
17077
  "z-ai/glm-5.2": {
17180
17078
  id: "z-ai/glm-5.2",
@@ -17185,9 +17083,9 @@ export const MODELS = {
17185
17083
  reasoning: true,
17186
17084
  input: ["text"],
17187
17085
  cost: {
17188
- input: 0.0341,
17189
- output: 7,
17190
- cacheRead: 0.032999999999999995,
17086
+ input: 0.006,
17087
+ output: 6.8,
17088
+ cacheRead: 0.005,
17191
17089
  cacheWrite: 0,
17192
17090
  },
17193
17091
  contextWindow: 1048576,
@@ -17202,9 +17100,9 @@ export const MODELS = {
17202
17100
  reasoning: true,
17203
17101
  input: ["text"],
17204
17102
  cost: {
17205
- input: 0.0699,
17206
- output: 1.4,
17207
- cacheRead: 0.0689,
17103
+ input: 0.08,
17104
+ output: 3.5999999999999996,
17105
+ cacheRead: 0.07600000000000001,
17208
17106
  cacheWrite: 0,
17209
17107
  },
17210
17108
  contextWindow: 1048576,
@@ -17219,13 +17117,13 @@ export const MODELS = {
17219
17117
  reasoning: true,
17220
17118
  input: ["text", "image"],
17221
17119
  cost: {
17222
- input: 0.15,
17223
- output: 0.5,
17224
- cacheRead: 0.03,
17120
+ input: 0.0285,
17121
+ output: 0.0923,
17122
+ cacheRead: 0.0285,
17225
17123
  cacheWrite: 0,
17226
17124
  },
17227
17125
  contextWindow: 1048576,
17228
- maxTokens: 943717,
17126
+ maxTokens: 131072,
17229
17127
  },
17230
17128
  "z-ai/glm-5.3-flash:batch": {
17231
17129
  id: "z-ai/glm-5.3-flash:batch",
@@ -17255,7 +17153,24 @@ export const MODELS = {
17255
17153
  cost: {
17256
17154
  input: 0.37,
17257
17155
  output: 1.25,
17258
- cacheRead: 0.09,
17156
+ cacheRead: 0.075,
17157
+ cacheWrite: 0,
17158
+ },
17159
+ contextWindow: 1048576,
17160
+ maxTokens: 131072,
17161
+ },
17162
+ "z-ai/glm-5.3-flashx-legal": {
17163
+ id: "z-ai/glm-5.3-flashx-legal",
17164
+ name: "Z.ai: GLM 5.3 FlashX Legal",
17165
+ api: "openai-completions",
17166
+ provider: "openrouter",
17167
+ baseUrl: "https://openrouter.ai/api/v1",
17168
+ reasoning: true,
17169
+ input: ["text", "image"],
17170
+ cost: {
17171
+ input: 0.37,
17172
+ output: 1.25,
17173
+ cacheRead: 0.075,
17259
17174
  cacheWrite: 0,
17260
17175
  },
17261
17176
  contextWindow: 1048576,
@@ -17389,9 +17304,9 @@ export const MODELS = {
17389
17304
  reasoning: true,
17390
17305
  input: ["text", "image"],
17391
17306
  cost: {
17392
- input: 0.022,
17393
- output: 1.331,
17394
- cacheRead: 0.013637,
17307
+ input: 0.035,
17308
+ output: 0.5,
17309
+ cacheRead: 0.005,
17395
17310
  cacheWrite: 0,
17396
17311
  },
17397
17312
  contextWindow: 1048576,
@@ -17406,9 +17321,9 @@ export const MODELS = {
17406
17321
  reasoning: true,
17407
17322
  input: ["text"],
17408
17323
  cost: {
17409
- input: 0.1259,
17324
+ input: 0.1035,
17410
17325
  output: 8,
17411
- cacheRead: 0.13,
17326
+ cacheRead: 0.09999999999999999,
17412
17327
  cacheWrite: 0,
17413
17328
  },
17414
17329
  contextWindow: 1048576,
@@ -17423,9 +17338,9 @@ export const MODELS = {
17423
17338
  reasoning: true,
17424
17339
  input: ["text"],
17425
17340
  cost: {
17426
- input: 0.0422,
17427
- output: 0.1694,
17428
- cacheRead: 0.028,
17341
+ input: 0.051399999999999994,
17342
+ output: 0.3,
17343
+ cacheRead: 0.035,
17429
17344
  cacheWrite: 0,
17430
17345
  },
17431
17346
  contextWindow: 1048576,
@@ -17474,9 +17389,9 @@ export const MODELS = {
17474
17389
  reasoning: true,
17475
17390
  input: ["text", "image"],
17476
17391
  cost: {
17477
- input: 0.16899999999999998,
17478
- output: 20,
17479
- cacheRead: 0.159,
17392
+ input: 0.14950000000000002,
17393
+ output: 7.15,
17394
+ cacheRead: 0.07475000000000001,
17480
17395
  cacheWrite: 0,
17481
17396
  },
17482
17397
  contextWindow: 1048576,
@@ -17610,9 +17525,9 @@ export const MODELS = {
17610
17525
  reasoning: true,
17611
17526
  input: ["text", "image"],
17612
17527
  cost: {
17613
- input: 0.034499999999999996,
17614
- output: 0.0823,
17615
- cacheRead: 0.03,
17528
+ input: 0.0285,
17529
+ output: 0.0923,
17530
+ cacheRead: 0.0285,
17616
17531
  cacheWrite: 0,
17617
17532
  },
17618
17533
  contextWindow: 1048576,
@@ -18657,23 +18572,6 @@ export const MODELS = {
18657
18572
  contextWindow: 1000000,
18658
18573
  maxTokens: 128000,
18659
18574
  },
18660
- "arcee-ai/trinity-large-thinking": {
18661
- id: "arcee-ai/trinity-large-thinking",
18662
- name: "Trinity Large Thinking",
18663
- api: "anthropic-messages",
18664
- provider: "vercel-ai-gateway",
18665
- baseUrl: "https://ai-gateway.vercel.sh",
18666
- reasoning: true,
18667
- input: ["text"],
18668
- cost: {
18669
- input: 0.25,
18670
- output: 0.7999999999999999,
18671
- cacheRead: 0.06,
18672
- cacheWrite: 0,
18673
- },
18674
- contextWindow: 262100,
18675
- maxTokens: 80000,
18676
- },
18677
18575
  "bytedance/seed-1.6": {
18678
18576
  id: "bytedance/seed-1.6",
18679
18577
  name: "Seed 1.6",
@@ -18921,9 +18819,9 @@ export const MODELS = {
18921
18819
  reasoning: true,
18922
18820
  input: ["text", "image"],
18923
18821
  cost: {
18924
- input: 0.075,
18925
- output: 0.3,
18926
- cacheRead: 0.0015,
18822
+ input: 0.3,
18823
+ output: 1.2,
18824
+ cacheRead: 0.007,
18927
18825
  cacheWrite: 0,
18928
18826
  },
18929
18827
  contextWindow: 1048576,
@@ -18938,13 +18836,13 @@ export const MODELS = {
18938
18836
  reasoning: true,
18939
18837
  input: ["text", "image"],
18940
18838
  cost: {
18941
- input: 0.075,
18942
- output: 0.3,
18943
- cacheRead: 0.0015,
18839
+ input: 0.3,
18840
+ output: 1.2,
18841
+ cacheRead: 0.007,
18944
18842
  cacheWrite: 0,
18945
18843
  },
18946
18844
  contextWindow: 1048576,
18947
- maxTokens: 1048576,
18845
+ maxTokens: 262144,
18948
18846
  },
18949
18847
  "deepseek/deepseek-v4.1-flash-1017-fast": {
18950
18848
  id: "deepseek/deepseek-v4.1-flash-1017-fast",
@@ -19558,6 +19456,23 @@ export const MODELS = {
19558
19456
  contextWindow: 262144,
19559
19457
  maxTokens: 32768,
19560
19458
  },
19459
+ "inclusionai/ling-3.1-flash-1019": {
19460
+ id: "inclusionai/ling-3.1-flash-1019",
19461
+ name: "Ling 3.1 Flash 1019",
19462
+ api: "anthropic-messages",
19463
+ provider: "vercel-ai-gateway",
19464
+ baseUrl: "https://ai-gateway.vercel.sh",
19465
+ reasoning: true,
19466
+ input: ["text"],
19467
+ cost: {
19468
+ input: 0,
19469
+ output: 0,
19470
+ cacheRead: 0,
19471
+ cacheWrite: 0,
19472
+ },
19473
+ contextWindow: 1000000,
19474
+ maxTokens: 262144,
19475
+ },
19561
19476
  "interfaze/interfaze-beta": {
19562
19477
  id: "interfaze/interfaze-beta",
19563
19478
  name: "Interfaze Beta",
@@ -19592,9 +19507,9 @@ export const MODELS = {
19592
19507
  contextWindow: 128000,
19593
19508
  maxTokens: 8200,
19594
19509
  },
19595
- "meituan/longcat-2.5-preview": {
19596
- id: "meituan/longcat-2.5-preview",
19597
- name: "LongCat 2.5 Preview",
19510
+ "meituan/longcat-2.5": {
19511
+ id: "meituan/longcat-2.5",
19512
+ name: "LongCat 2.5",
19598
19513
  api: "anthropic-messages",
19599
19514
  provider: "vercel-ai-gateway",
19600
19515
  baseUrl: "https://ai-gateway.vercel.sh",
@@ -20077,12 +19992,12 @@ export const MODELS = {
20077
19992
  reasoning: true,
20078
19993
  input: ["text", "image"],
20079
19994
  cost: {
20080
- input: 1.3599999999999999,
20081
- output: 4.18,
20082
- cacheRead: 0.14,
19995
+ input: 0.7,
19996
+ output: 2.0999999999999996,
19997
+ cacheRead: 0.07,
20083
19998
  cacheWrite: 0,
20084
19999
  },
20085
- contextWindow: 524288,
20000
+ contextWindow: 1048576,
20086
20001
  maxTokens: 262144,
20087
20002
  },
20088
20003
  "mistral/mistral-medium-3.5": {
@@ -21853,39 +21768,39 @@ export const MODELS = {
21853
21768
  contextWindow: 200000,
21854
21769
  maxTokens: 100000,
21855
21770
  },
21856
- "poolside/laguna-s-2.1": {
21857
- id: "poolside/laguna-s-2.1",
21858
- name: "Laguna S 2.1",
21771
+ "particle/particle-1": {
21772
+ id: "particle/particle-1",
21773
+ name: "Particle-1",
21859
21774
  api: "anthropic-messages",
21860
21775
  provider: "vercel-ai-gateway",
21861
21776
  baseUrl: "https://ai-gateway.vercel.sh",
21862
21777
  reasoning: true,
21863
- input: ["text"],
21778
+ input: ["text", "image"],
21864
21779
  cost: {
21865
- input: 0.09,
21866
- output: 0.18,
21867
- cacheRead: 0.009,
21780
+ input: 0.125,
21781
+ output: 0.5,
21782
+ cacheRead: 0.015,
21868
21783
  cacheWrite: 0,
21869
21784
  },
21870
- contextWindow: 1000000,
21785
+ contextWindow: 1048576,
21871
21786
  maxTokens: 131072,
21872
21787
  },
21873
- "poolside/laguna-s-2.1-free": {
21874
- id: "poolside/laguna-s-2.1-free",
21875
- name: "Laguna S 2.1 Free",
21788
+ "poolside/laguna-s-2.1": {
21789
+ id: "poolside/laguna-s-2.1",
21790
+ name: "Laguna S 2.1",
21876
21791
  api: "anthropic-messages",
21877
21792
  provider: "vercel-ai-gateway",
21878
21793
  baseUrl: "https://ai-gateway.vercel.sh",
21879
21794
  reasoning: true,
21880
21795
  input: ["text"],
21881
21796
  cost: {
21882
- input: 0,
21883
- output: 0,
21884
- cacheRead: 0,
21797
+ input: 0.09,
21798
+ output: 0.18,
21799
+ cacheRead: 0.009,
21885
21800
  cacheWrite: 0,
21886
21801
  },
21887
- contextWindow: 256000,
21888
- maxTokens: 32768,
21802
+ contextWindow: 1000000,
21803
+ maxTokens: 131072,
21889
21804
  },
21890
21805
  "quiverai/arrow-2": {
21891
21806
  id: "quiverai/arrow-2",
@@ -22414,40 +22329,6 @@ export const MODELS = {
22414
22329
  contextWindow: 1000000,
22415
22330
  maxTokens: 1000000,
22416
22331
  },
22417
- "xiaomi/mimo-v2.5": {
22418
- id: "xiaomi/mimo-v2.5",
22419
- name: "MiMo M2.5",
22420
- api: "anthropic-messages",
22421
- provider: "vercel-ai-gateway",
22422
- baseUrl: "https://ai-gateway.vercel.sh",
22423
- reasoning: true,
22424
- input: ["text", "image"],
22425
- cost: {
22426
- input: 0.119,
22427
- output: 0.238,
22428
- cacheRead: 0.0025,
22429
- cacheWrite: 0,
22430
- },
22431
- contextWindow: 1050000,
22432
- maxTokens: 131100,
22433
- },
22434
- "xiaomi/mimo-v2.5-pro": {
22435
- id: "xiaomi/mimo-v2.5-pro",
22436
- name: "MiMo V2.5 Pro",
22437
- api: "anthropic-messages",
22438
- provider: "vercel-ai-gateway",
22439
- baseUrl: "https://ai-gateway.vercel.sh",
22440
- reasoning: true,
22441
- input: ["text"],
22442
- cost: {
22443
- input: 0.3045,
22444
- output: 0.609,
22445
- cacheRead: 0.0028,
22446
- cacheWrite: 0,
22447
- },
22448
- contextWindow: 1050000,
22449
- maxTokens: 131000,
22450
- },
22451
22332
  "xiaomi/mimo-v2.6-flash": {
22452
22333
  id: "xiaomi/mimo-v2.6-flash",
22453
22334
  name: "MiMo V2.6 Flash",
@@ -22712,13 +22593,13 @@ export const MODELS = {
22712
22593
  reasoning: true,
22713
22594
  input: ["text"],
22714
22595
  cost: {
22715
- input: 1.2,
22716
- output: 4,
22717
- cacheRead: 0.19999999999999998,
22596
+ input: 1.4,
22597
+ output: 4.4,
22598
+ cacheRead: 0.14,
22718
22599
  cacheWrite: 0,
22719
22600
  },
22720
22601
  contextWindow: 1000000,
22721
- maxTokens: 262144,
22602
+ maxTokens: 1000000,
22722
22603
  },
22723
22604
  "zai/glm-5.3-fast": {
22724
22605
  id: "zai/glm-5.3-fast",
@@ -22771,6 +22652,23 @@ export const MODELS = {
22771
22652
  contextWindow: 1000000,
22772
22653
  maxTokens: 131072,
22773
22654
  },
22655
+ "zai/glm-5.3-flashx-legal": {
22656
+ id: "zai/glm-5.3-flashx-legal",
22657
+ name: "GLM 5.3 FlashX Legal",
22658
+ api: "anthropic-messages",
22659
+ provider: "vercel-ai-gateway",
22660
+ baseUrl: "https://ai-gateway.vercel.sh",
22661
+ reasoning: true,
22662
+ input: ["text", "image"],
22663
+ cost: {
22664
+ input: 0.37,
22665
+ output: 1.25,
22666
+ cacheRead: 0.075,
22667
+ cacheWrite: 0,
22668
+ },
22669
+ contextWindow: 1000000,
22670
+ maxTokens: 131072,
22671
+ },
22774
22672
  "zai/glm-5v-turbo": {
22775
22673
  id: "zai/glm-5v-turbo",
22776
22674
  name: "GLM 5V Turbo",