omk-ai 1.2.4 → 1.3.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (39) hide show
  1. package/CHANGELOG.md +27 -0
  2. package/dist/image-models.generated.d.ts +45 -0
  3. package/dist/image-models.generated.d.ts.map +1 -1
  4. package/dist/image-models.generated.js +45 -0
  5. package/dist/image-models.generated.js.map +1 -1
  6. package/dist/models.generated.d.ts +1509 -864
  7. package/dist/models.generated.d.ts.map +1 -1
  8. package/dist/models.generated.js +1215 -709
  9. package/dist/models.generated.js.map +1 -1
  10. package/dist/providers/cursor-connect.d.ts +3 -0
  11. package/dist/providers/cursor-connect.d.ts.map +1 -0
  12. package/dist/providers/cursor-connect.js +9 -0
  13. package/dist/providers/cursor-connect.js.map +1 -0
  14. package/dist/providers/cursor.d.ts.map +1 -1
  15. package/dist/providers/cursor.js +17 -13
  16. package/dist/providers/cursor.js.map +1 -1
  17. package/dist/stream.d.ts.map +1 -1
  18. package/dist/stream.js +6 -0
  19. package/dist/stream.js.map +1 -1
  20. package/dist/utils/provider-retry-delay.d.ts +4 -0
  21. package/dist/utils/provider-retry-delay.d.ts.map +1 -0
  22. package/dist/utils/provider-retry-delay.js +53 -0
  23. package/dist/utils/provider-retry-delay.js.map +1 -0
  24. package/dist/utils/provider-retry-sleep.d.ts +6 -0
  25. package/dist/utils/provider-retry-sleep.d.ts.map +1 -0
  26. package/dist/utils/provider-retry-sleep.js +51 -0
  27. package/dist/utils/provider-retry-sleep.js.map +1 -0
  28. package/dist/utils/provider-retry.d.ts.map +1 -1
  29. package/dist/utils/provider-retry.js +6 -44
  30. package/dist/utils/provider-retry.js.map +1 -1
  31. package/dist/utils/retry-backoff.d.ts +8 -0
  32. package/dist/utils/retry-backoff.d.ts.map +1 -0
  33. package/dist/utils/retry-backoff.js +19 -0
  34. package/dist/utils/retry-backoff.js.map +1 -0
  35. package/dist/utils/retry.d.ts +11 -3
  36. package/dist/utils/retry.d.ts.map +1 -1
  37. package/dist/utils/retry.js +13 -5
  38. package/dist/utils/retry.js.map +1 -1
  39. package/package.json +1 -1
@@ -1190,6 +1190,40 @@ export const MODELS = {
1190
1190
  contextWindow: 1050000,
1191
1191
  maxTokens: 128000,
1192
1192
  },
1193
+ "global.openai.gpt-6-luna": {
1194
+ id: "global.openai.gpt-6-luna",
1195
+ name: "GPT-6 Luna (Global)",
1196
+ api: "bedrock-converse-stream",
1197
+ provider: "amazon-bedrock",
1198
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1199
+ reasoning: true,
1200
+ input: ["text", "image"],
1201
+ cost: {
1202
+ input: 0.1,
1203
+ output: 0.5,
1204
+ cacheRead: 0.01,
1205
+ cacheWrite: 0.125,
1206
+ },
1207
+ contextWindow: 1050000,
1208
+ maxTokens: 128000,
1209
+ },
1210
+ "global.openai.gpt-6-sol": {
1211
+ id: "global.openai.gpt-6-sol",
1212
+ name: "GPT-6 Sol (Global)",
1213
+ api: "bedrock-converse-stream",
1214
+ provider: "amazon-bedrock",
1215
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1216
+ reasoning: true,
1217
+ input: ["text", "image"],
1218
+ cost: {
1219
+ input: 2,
1220
+ output: 10,
1221
+ cacheRead: 0.2,
1222
+ cacheWrite: 2.5,
1223
+ },
1224
+ contextWindow: 1050000,
1225
+ maxTokens: 128000,
1226
+ },
1193
1227
  "global.xai.grok-4.6": {
1194
1228
  id: "global.xai.grok-4.6",
1195
1229
  name: "Grok 4.6 (Global)",
@@ -1208,6 +1242,24 @@ export const MODELS = {
1208
1242
  contextWindow: 500000,
1209
1243
  maxTokens: 500000,
1210
1244
  },
1245
+ "global.xai.grok-4.7": {
1246
+ id: "global.xai.grok-4.7",
1247
+ name: "Grok 4.7 (Global)",
1248
+ api: "bedrock-converse-stream",
1249
+ provider: "amazon-bedrock",
1250
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
1251
+ reasoning: true,
1252
+ thinkingLevelMap: { "xhigh": "xhigh" },
1253
+ input: ["text", "image"],
1254
+ cost: {
1255
+ input: 2,
1256
+ output: 6,
1257
+ cacheRead: 0.5,
1258
+ cacheWrite: 0,
1259
+ },
1260
+ contextWindow: 500000,
1261
+ maxTokens: 500000,
1262
+ },
1211
1263
  "google.gemma-4-26b-a4b": {
1212
1264
  id: "google.gemma-4-26b-a4b",
1213
1265
  name: "Gemma 4 26B A4B IT",
@@ -1569,7 +1621,7 @@ export const MODELS = {
1569
1621
  cacheRead: 0,
1570
1622
  cacheWrite: 0,
1571
1623
  },
1572
- contextWindow: 204800,
1624
+ contextWindow: 196608,
1573
1625
  maxTokens: 131072,
1574
1626
  },
1575
1627
  "minimax.minimax-m2.5": {
@@ -1603,7 +1655,7 @@ export const MODELS = {
1603
1655
  cacheRead: 0,
1604
1656
  cacheWrite: 0,
1605
1657
  },
1606
- contextWindow: 256000,
1658
+ contextWindow: 262144,
1607
1659
  maxTokens: 8192,
1608
1660
  },
1609
1661
  "mistral.magistral-small-2509": {
@@ -1688,7 +1740,7 @@ export const MODELS = {
1688
1740
  cacheRead: 0,
1689
1741
  cacheWrite: 0,
1690
1742
  },
1691
- contextWindow: 256000,
1743
+ contextWindow: 262144,
1692
1744
  maxTokens: 8192,
1693
1745
  },
1694
1746
  "mistral.pixtral-large-2502-v1:0": {
@@ -1756,7 +1808,7 @@ export const MODELS = {
1756
1808
  cacheRead: 0,
1757
1809
  cacheWrite: 0,
1758
1810
  },
1759
- contextWindow: 262143,
1811
+ contextWindow: 262144,
1760
1812
  maxTokens: 16000,
1761
1813
  },
1762
1814
  "moonshotai.kimi-k2.5": {
@@ -1773,7 +1825,7 @@ export const MODELS = {
1773
1825
  cacheRead: 0,
1774
1826
  cacheWrite: 0,
1775
1827
  },
1776
- contextWindow: 262143,
1828
+ contextWindow: 262144,
1777
1829
  maxTokens: 16384,
1778
1830
  },
1779
1831
  "nvidia.nemotron-nano-12b-v2": {
@@ -1790,7 +1842,7 @@ export const MODELS = {
1790
1842
  cacheRead: 0,
1791
1843
  cacheWrite: 0,
1792
1844
  },
1793
- contextWindow: 128000,
1845
+ contextWindow: 131072,
1794
1846
  maxTokens: 8192,
1795
1847
  },
1796
1848
  "nvidia.nemotron-nano-3-30b": {
@@ -1859,7 +1911,7 @@ export const MODELS = {
1859
1911
  cacheRead: 0.275,
1860
1912
  cacheWrite: 0,
1861
1913
  },
1862
- contextWindow: 272000,
1914
+ contextWindow: 1000000,
1863
1915
  maxTokens: 128000,
1864
1916
  },
1865
1917
  "openai.gpt-5.5": {
@@ -1877,7 +1929,7 @@ export const MODELS = {
1877
1929
  cacheRead: 0.55,
1878
1930
  cacheWrite: 0,
1879
1931
  },
1880
- contextWindow: 272000,
1932
+ contextWindow: 1000000,
1881
1933
  maxTokens: 128000,
1882
1934
  },
1883
1935
  "openai.gpt-5.6-luna": {
@@ -1951,6 +2003,40 @@ export const MODELS = {
1951
2003
  contextWindow: 1050000,
1952
2004
  maxTokens: 128000,
1953
2005
  },
2006
+ "openai.gpt-6-luna": {
2007
+ id: "openai.gpt-6-luna",
2008
+ name: "GPT-6 Luna",
2009
+ api: "bedrock-converse-stream",
2010
+ provider: "amazon-bedrock",
2011
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2012
+ reasoning: true,
2013
+ input: ["text", "image"],
2014
+ cost: {
2015
+ input: 0.11,
2016
+ output: 0.55,
2017
+ cacheRead: 0.011,
2018
+ cacheWrite: 0.1375,
2019
+ },
2020
+ contextWindow: 1050000,
2021
+ maxTokens: 128000,
2022
+ },
2023
+ "openai.gpt-6-sol": {
2024
+ id: "openai.gpt-6-sol",
2025
+ name: "GPT-6 Sol",
2026
+ api: "bedrock-converse-stream",
2027
+ provider: "amazon-bedrock",
2028
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2029
+ reasoning: true,
2030
+ input: ["text", "image"],
2031
+ cost: {
2032
+ input: 2.2,
2033
+ output: 11,
2034
+ cacheRead: 0.22,
2035
+ cacheWrite: 2.75,
2036
+ },
2037
+ contextWindow: 1050000,
2038
+ maxTokens: 128000,
2039
+ },
1954
2040
  "openai.gpt-oss-120b": {
1955
2041
  id: "openai.gpt-oss-120b",
1956
2042
  name: "gpt-oss-120b",
@@ -1965,8 +2051,8 @@ export const MODELS = {
1965
2051
  cacheRead: 0,
1966
2052
  cacheWrite: 0,
1967
2053
  },
1968
- contextWindow: 128000,
1969
- maxTokens: 16384,
2054
+ contextWindow: 131072,
2055
+ maxTokens: 131072,
1970
2056
  },
1971
2057
  "openai.gpt-oss-120b-1:0": {
1972
2058
  id: "openai.gpt-oss-120b-1:0",
@@ -1982,8 +2068,8 @@ export const MODELS = {
1982
2068
  cacheRead: 0,
1983
2069
  cacheWrite: 0,
1984
2070
  },
1985
- contextWindow: 128000,
1986
- maxTokens: 16384,
2071
+ contextWindow: 131072,
2072
+ maxTokens: 128000,
1987
2073
  },
1988
2074
  "openai.gpt-oss-20b": {
1989
2075
  id: "openai.gpt-oss-20b",
@@ -1999,8 +2085,8 @@ export const MODELS = {
1999
2085
  cacheRead: 0,
2000
2086
  cacheWrite: 0,
2001
2087
  },
2002
- contextWindow: 128000,
2003
- maxTokens: 16384,
2088
+ contextWindow: 131072,
2089
+ maxTokens: 131072,
2004
2090
  },
2005
2091
  "openai.gpt-oss-20b-1:0": {
2006
2092
  id: "openai.gpt-oss-20b-1:0",
@@ -2016,8 +2102,8 @@ export const MODELS = {
2016
2102
  cacheRead: 0,
2017
2103
  cacheWrite: 0,
2018
2104
  },
2019
- contextWindow: 128000,
2020
- maxTokens: 16384,
2105
+ contextWindow: 131072,
2106
+ maxTokens: 128000,
2021
2107
  },
2022
2108
  "openai.gpt-oss-safeguard-120b": {
2023
2109
  id: "openai.gpt-oss-safeguard-120b",
@@ -2711,6 +2797,40 @@ export const MODELS = {
2711
2797
  contextWindow: 1050000,
2712
2798
  maxTokens: 128000,
2713
2799
  },
2800
+ "us.openai.gpt-6-luna": {
2801
+ id: "us.openai.gpt-6-luna",
2802
+ name: "GPT-6 Luna (US)",
2803
+ api: "bedrock-converse-stream",
2804
+ provider: "amazon-bedrock",
2805
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2806
+ reasoning: true,
2807
+ input: ["text", "image"],
2808
+ cost: {
2809
+ input: 0.11,
2810
+ output: 0.55,
2811
+ cacheRead: 0.011,
2812
+ cacheWrite: 0.1375,
2813
+ },
2814
+ contextWindow: 1050000,
2815
+ maxTokens: 128000,
2816
+ },
2817
+ "us.openai.gpt-6-sol": {
2818
+ id: "us.openai.gpt-6-sol",
2819
+ name: "GPT-6 Sol (US)",
2820
+ api: "bedrock-converse-stream",
2821
+ provider: "amazon-bedrock",
2822
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2823
+ reasoning: true,
2824
+ input: ["text", "image"],
2825
+ cost: {
2826
+ input: 2.2,
2827
+ output: 11,
2828
+ cacheRead: 0.22,
2829
+ cacheWrite: 2.75,
2830
+ },
2831
+ contextWindow: 1050000,
2832
+ maxTokens: 128000,
2833
+ },
2714
2834
  "us.writer.palmyra-x4-v1:0": {
2715
2835
  id: "us.writer.palmyra-x4-v1:0",
2716
2836
  name: "Palmyra X4 (US)",
@@ -2763,6 +2883,24 @@ export const MODELS = {
2763
2883
  contextWindow: 500000,
2764
2884
  maxTokens: 500000,
2765
2885
  },
2886
+ "us.xai.grok-4.7": {
2887
+ id: "us.xai.grok-4.7",
2888
+ name: "Grok 4.7 (US)",
2889
+ api: "bedrock-converse-stream",
2890
+ provider: "amazon-bedrock",
2891
+ baseUrl: "https://bedrock-runtime.us-east-1.amazonaws.com",
2892
+ reasoning: true,
2893
+ thinkingLevelMap: { "xhigh": "xhigh" },
2894
+ input: ["text", "image"],
2895
+ cost: {
2896
+ input: 2.2,
2897
+ output: 6.6,
2898
+ cacheRead: 0.55,
2899
+ cacheWrite: 0,
2900
+ },
2901
+ contextWindow: 500000,
2902
+ maxTokens: 500000,
2903
+ },
2766
2904
  "writer.palmyra-x4-v1:0": {
2767
2905
  id: "writer.palmyra-x4-v1:0",
2768
2906
  name: "Palmyra X4",
@@ -2846,7 +2984,7 @@ export const MODELS = {
2846
2984
  cacheRead: 0,
2847
2985
  cacheWrite: 0,
2848
2986
  },
2849
- contextWindow: 204800,
2987
+ contextWindow: 202752,
2850
2988
  maxTokens: 131072,
2851
2989
  },
2852
2990
  "zai.glm-4.7-flash": {
@@ -2863,7 +3001,7 @@ export const MODELS = {
2863
3001
  cacheRead: 0,
2864
3002
  cacheWrite: 0,
2865
3003
  },
2866
- contextWindow: 200000,
3004
+ contextWindow: 202752,
2867
3005
  maxTokens: 131072,
2868
3006
  },
2869
3007
  "zai.glm-5": {
@@ -3872,6 +4010,58 @@ export const MODELS = {
3872
4010
  contextWindow: 1050000,
3873
4011
  maxTokens: 128000,
3874
4012
  },
4013
+ "gpt-6.1-sol": {
4014
+ id: "gpt-6.1-sol",
4015
+ name: "GPT-6.1 Sol",
4016
+ api: "azure-openai-responses",
4017
+ provider: "azure-openai-responses",
4018
+ baseUrl: "",
4019
+ reasoning: true,
4020
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
4021
+ input: ["text", "image"],
4022
+ cost: {
4023
+ input: 2,
4024
+ output: 10,
4025
+ cacheRead: 0.1,
4026
+ cacheWrite: 2.5,
4027
+ },
4028
+ contextWindow: 1050000,
4029
+ maxTokens: 128000,
4030
+ },
4031
+ "gpt-daybreak-blue-latest": {
4032
+ id: "gpt-daybreak-blue-latest",
4033
+ name: "Daybreak Blue",
4034
+ api: "azure-openai-responses",
4035
+ provider: "azure-openai-responses",
4036
+ baseUrl: "",
4037
+ reasoning: true,
4038
+ input: ["text", "image"],
4039
+ cost: {
4040
+ input: 4,
4041
+ output: 20,
4042
+ cacheRead: 0.4,
4043
+ cacheWrite: 5,
4044
+ },
4045
+ contextWindow: 1050000,
4046
+ maxTokens: 128000,
4047
+ },
4048
+ "gpt-daybreak-red-latest": {
4049
+ id: "gpt-daybreak-red-latest",
4050
+ name: "Daybreak Red",
4051
+ api: "azure-openai-responses",
4052
+ provider: "azure-openai-responses",
4053
+ baseUrl: "",
4054
+ reasoning: true,
4055
+ input: ["text", "image"],
4056
+ cost: {
4057
+ input: 12.5,
4058
+ output: 75,
4059
+ cacheRead: 1.25,
4060
+ cacheWrite: 15.625,
4061
+ },
4062
+ contextWindow: 400000,
4063
+ maxTokens: 128000,
4064
+ },
3875
4065
  "gpt-realtime-2.1": {
3876
4066
  id: "gpt-realtime-2.1",
3877
4067
  name: "GPT-Realtime-2.1",
@@ -4025,8 +4215,8 @@ export const MODELS = {
4025
4215
  cacheRead: 0,
4026
4216
  cacheWrite: 0,
4027
4217
  },
4028
- contextWindow: 65536,
4029
- maxTokens: 32768,
4218
+ contextWindow: 131072,
4219
+ maxTokens: 40960,
4030
4220
  },
4031
4221
  },
4032
4222
  "cloudflare-ai-gateway": {
@@ -4178,6 +4368,25 @@ export const MODELS = {
4178
4368
  contextWindow: 1000000,
4179
4369
  maxTokens: 128000,
4180
4370
  },
4371
+ "claude-opus-5.5": {
4372
+ id: "claude-opus-5.5",
4373
+ name: "Claude Opus 5.5",
4374
+ api: "anthropic-messages",
4375
+ provider: "cloudflare-ai-gateway",
4376
+ baseUrl: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/anthropic",
4377
+ compat: { "forceAdaptiveThinking": true },
4378
+ reasoning: true,
4379
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
4380
+ input: ["text", "image"],
4381
+ cost: {
4382
+ input: 4,
4383
+ output: 20,
4384
+ cacheRead: 0.2,
4385
+ cacheWrite: 5,
4386
+ },
4387
+ contextWindow: 1000000,
4388
+ maxTokens: 128000,
4389
+ },
4181
4390
  "claude-sonnet-4.5": {
4182
4391
  id: "claude-sonnet-4.5",
4183
4392
  name: "Claude Sonnet 4.5 (latest)",
@@ -4192,7 +4401,7 @@ export const MODELS = {
4192
4401
  cacheRead: 0.3,
4193
4402
  cacheWrite: 3.75,
4194
4403
  },
4195
- contextWindow: 1000000,
4404
+ contextWindow: 200000,
4196
4405
  maxTokens: 64000,
4197
4406
  },
4198
4407
  "claude-sonnet-4.6": {
@@ -4569,6 +4778,42 @@ export const MODELS = {
4569
4778
  contextWindow: 1050000,
4570
4779
  maxTokens: 128000,
4571
4780
  },
4781
+ "gpt-6-luna": {
4782
+ id: "gpt-6-luna",
4783
+ name: "GPT-6 Luna",
4784
+ api: "openai-responses",
4785
+ provider: "cloudflare-ai-gateway",
4786
+ baseUrl: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai",
4787
+ reasoning: true,
4788
+ thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
4789
+ input: ["text", "image"],
4790
+ cost: {
4791
+ input: 0.1,
4792
+ output: 0.5,
4793
+ cacheRead: 0.01,
4794
+ cacheWrite: 0.125,
4795
+ },
4796
+ contextWindow: 1050000,
4797
+ maxTokens: 128000,
4798
+ },
4799
+ "gpt-6-sol": {
4800
+ id: "gpt-6-sol",
4801
+ name: "GPT-6 Sol",
4802
+ api: "openai-responses",
4803
+ provider: "cloudflare-ai-gateway",
4804
+ baseUrl: "https://gateway.ai.cloudflare.com/v1/{CLOUDFLARE_ACCOUNT_ID}/{CLOUDFLARE_GATEWAY_ID}/openai",
4805
+ reasoning: true,
4806
+ thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
4807
+ input: ["text", "image"],
4808
+ cost: {
4809
+ input: 2,
4810
+ output: 10,
4811
+ cacheRead: 0.2,
4812
+ cacheWrite: 2.5,
4813
+ },
4814
+ contextWindow: 1050000,
4815
+ maxTokens: 128000,
4816
+ },
4572
4817
  "o3": {
4573
4818
  id: "o3",
4574
4819
  name: "o3",
@@ -4638,7 +4883,7 @@ export const MODELS = {
4638
4883
  cacheRead: 0.014,
4639
4884
  cacheWrite: 0,
4640
4885
  },
4641
- contextWindow: 1310720,
4886
+ contextWindow: 1048576,
4642
4887
  maxTokens: 1048576,
4643
4888
  },
4644
4889
  "@cf/deepseek-ai/deepseek-v4-pro-0813": {
@@ -4931,7 +5176,7 @@ export const MODELS = {
4931
5176
  cacheRead: 0.26,
4932
5177
  cacheWrite: 0,
4933
5178
  },
4934
- contextWindow: 1310720,
5179
+ contextWindow: 1048576,
4935
5180
  maxTokens: 1048576,
4936
5181
  },
4937
5182
  "@cf/zai-org/glm-5.3-flash": {
@@ -4950,7 +5195,7 @@ export const MODELS = {
4950
5195
  cacheRead: 0.03,
4951
5196
  cacheWrite: 0,
4952
5197
  },
4953
- contextWindow: 1310720,
5198
+ contextWindow: 1048576,
4954
5199
  maxTokens: 1048576,
4955
5200
  },
4956
5201
  },
@@ -12669,15 +12914,15 @@ export const MODELS = {
12669
12914
  },
12670
12915
  },
12671
12916
  "fireworks": {
12672
- "accounts/fireworks/models/deepseek-v4-flash-0731": {
12673
- id: "accounts/fireworks/models/deepseek-v4-flash-0731",
12674
- name: "DeepSeek V4 Flash 0731",
12917
+ "accounts/fireworks/models/deepseek-v4p1-flash": {
12918
+ id: "accounts/fireworks/models/deepseek-v4p1-flash",
12919
+ name: "DeepSeek V4.1 Flash",
12675
12920
  api: "anthropic-messages",
12676
12921
  provider: "fireworks",
12677
12922
  baseUrl: "https://api.fireworks.ai/inference",
12678
12923
  compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
12679
12924
  reasoning: true,
12680
- input: ["text"],
12925
+ input: ["text", "image"],
12681
12926
  cost: {
12682
12927
  input: 0.22,
12683
12928
  output: 0.66,
@@ -12687,9 +12932,9 @@ export const MODELS = {
12687
12932
  contextWindow: 1000000,
12688
12933
  maxTokens: 384000,
12689
12934
  },
12690
- "accounts/fireworks/models/deepseek-v4-flash-vision-exp": {
12691
- id: "accounts/fireworks/models/deepseek-v4-flash-vision-exp",
12692
- name: "DeepSeek V4 Flash Vision Exp",
12935
+ "accounts/fireworks/models/ember-1": {
12936
+ id: "accounts/fireworks/models/ember-1",
12937
+ name: "Ember-1",
12693
12938
  api: "anthropic-messages",
12694
12939
  provider: "fireworks",
12695
12940
  baseUrl: "https://api.fireworks.ai/inference",
@@ -12697,85 +12942,12 @@ export const MODELS = {
12697
12942
  reasoning: true,
12698
12943
  input: ["text", "image"],
12699
12944
  cost: {
12700
- input: 0.22,
12701
- output: 0.66,
12702
- cacheRead: 0.007,
12703
- cacheWrite: 0,
12704
- },
12705
- contextWindow: 1000000,
12706
- maxTokens: 384000,
12707
- },
12708
- "accounts/fireworks/models/deepseek-v4-pro": {
12709
- id: "accounts/fireworks/models/deepseek-v4-pro",
12710
- name: "DeepSeek V4 Pro",
12711
- api: "anthropic-messages",
12712
- provider: "fireworks",
12713
- baseUrl: "https://api.fireworks.ai/inference",
12714
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
12715
- reasoning: true,
12716
- input: ["text"],
12717
- cost: {
12718
- input: 1.2,
12719
- output: 1.2,
12720
- cacheRead: 0.6,
12721
- cacheWrite: 0,
12722
- },
12723
- contextWindow: 1000000,
12724
- maxTokens: 384000,
12725
- },
12726
- "accounts/fireworks/models/deepseek-v4-pro-0813": {
12727
- id: "accounts/fireworks/models/deepseek-v4-pro-0813",
12728
- name: "DeepSeek V4 Pro 0813",
12729
- api: "anthropic-messages",
12730
- provider: "fireworks",
12731
- baseUrl: "https://api.fireworks.ai/inference",
12732
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
12733
- reasoning: true,
12734
- input: ["text"],
12735
- cost: {
12736
- input: 1.32,
12737
- output: 3.96,
12738
- cacheRead: 0.044,
12739
- cacheWrite: 0,
12740
- },
12741
- contextWindow: 1000000,
12742
- maxTokens: 384000,
12743
- },
12744
- "accounts/fireworks/models/deepseek-v4p1-flash": {
12745
- id: "accounts/fireworks/models/deepseek-v4p1-flash",
12746
- name: "DeepSeek V4.1 Flash",
12747
- api: "anthropic-messages",
12748
- provider: "fireworks",
12749
- baseUrl: "https://api.fireworks.ai/inference",
12750
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
12751
- reasoning: true,
12752
- input: ["text", "image"],
12753
- cost: {
12754
- input: 0.22,
12755
- output: 0.66,
12756
- cacheRead: 0.007,
12757
- cacheWrite: 0,
12758
- },
12759
- contextWindow: 1000000,
12760
- maxTokens: 384000,
12761
- },
12762
- "accounts/fireworks/models/glm-5p2": {
12763
- id: "accounts/fireworks/models/glm-5p2",
12764
- name: "GLM 5.2",
12765
- api: "anthropic-messages",
12766
- provider: "fireworks",
12767
- baseUrl: "https://api.fireworks.ai/inference",
12768
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
12769
- reasoning: true,
12770
- thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
12771
- input: ["text"],
12772
- cost: {
12773
- input: 1.4,
12774
- output: 4.4,
12775
- cacheRead: 0.14,
12945
+ input: 3,
12946
+ output: 15,
12947
+ cacheRead: 0.3,
12776
12948
  cacheWrite: 0,
12777
12949
  },
12778
- contextWindow: 1048575,
12950
+ contextWindow: 1048576,
12779
12951
  maxTokens: 131072,
12780
12952
  },
12781
12953
  "accounts/fireworks/models/glm-5p3": {
@@ -12852,42 +13024,6 @@ export const MODELS = {
12852
13024
  contextWindow: 1048576,
12853
13025
  maxTokens: 1048576,
12854
13026
  },
12855
- "accounts/fireworks/models/kimi-k2p6": {
12856
- id: "accounts/fireworks/models/kimi-k2p6",
12857
- name: "Kimi K2.6",
12858
- api: "anthropic-messages",
12859
- provider: "fireworks",
12860
- baseUrl: "https://api.fireworks.ai/inference",
12861
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
12862
- reasoning: true,
12863
- input: ["text", "image"],
12864
- cost: {
12865
- input: 0.95,
12866
- output: 4,
12867
- cacheRead: 0.16,
12868
- cacheWrite: 0,
12869
- },
12870
- contextWindow: 262000,
12871
- maxTokens: 262000,
12872
- },
12873
- "accounts/fireworks/models/kimi-k2p7-code": {
12874
- id: "accounts/fireworks/models/kimi-k2p7-code",
12875
- name: "Kimi K2.7 Code",
12876
- api: "anthropic-messages",
12877
- provider: "fireworks",
12878
- baseUrl: "https://api.fireworks.ai/inference",
12879
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
12880
- reasoning: true,
12881
- input: ["text", "image"],
12882
- cost: {
12883
- input: 0.95,
12884
- output: 4,
12885
- cacheRead: 0.19,
12886
- cacheWrite: 0,
12887
- },
12888
- contextWindow: 262000,
12889
- maxTokens: 262000,
12890
- },
12891
13027
  "accounts/fireworks/models/kimi-k3": {
12892
13028
  id: "accounts/fireworks/models/kimi-k3",
12893
13029
  name: "Kimi K3",
@@ -12906,24 +13042,6 @@ export const MODELS = {
12906
13042
  contextWindow: 1048576,
12907
13043
  maxTokens: 131072,
12908
13044
  },
12909
- "accounts/fireworks/models/minimax-m2p7": {
12910
- id: "accounts/fireworks/models/minimax-m2p7",
12911
- name: "MiniMax-M2.7",
12912
- api: "anthropic-messages",
12913
- provider: "fireworks",
12914
- baseUrl: "https://api.fireworks.ai/inference",
12915
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
12916
- reasoning: true,
12917
- input: ["text"],
12918
- cost: {
12919
- input: 1.2,
12920
- output: 1.2,
12921
- cacheRead: 0.6,
12922
- cacheWrite: 0,
12923
- },
12924
- contextWindow: 196608,
12925
- maxTokens: 131072,
12926
- },
12927
13045
  "accounts/fireworks/models/minimax-m3": {
12928
13046
  id: "accounts/fireworks/models/minimax-m3",
12929
13047
  name: "MiniMax-M3",
@@ -12942,24 +13060,6 @@ export const MODELS = {
12942
13060
  contextWindow: 512000,
12943
13061
  maxTokens: 512000,
12944
13062
  },
12945
- "accounts/fireworks/models/muse-glimmer-30b": {
12946
- id: "accounts/fireworks/models/muse-glimmer-30b",
12947
- name: "Muse Glimmer 30B",
12948
- api: "anthropic-messages",
12949
- provider: "fireworks",
12950
- baseUrl: "https://api.fireworks.ai/inference",
12951
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
12952
- reasoning: true,
12953
- input: ["text", "image"],
12954
- cost: {
12955
- input: 0.35,
12956
- output: 1.5,
12957
- cacheRead: 0.04,
12958
- cacheWrite: 0,
12959
- },
12960
- contextWindow: 131072,
12961
- maxTokens: 131072,
12962
- },
12963
13063
  "accounts/fireworks/models/nemotron-3-ultra-nvfp4": {
12964
13064
  id: "accounts/fireworks/models/nemotron-3-ultra-nvfp4",
12965
13065
  name: "Nemotron 3 Ultra 550B A55B",
@@ -12996,25 +13096,6 @@ export const MODELS = {
12996
13096
  contextWindow: 262144,
12997
13097
  maxTokens: 262144,
12998
13098
  },
12999
- "accounts/fireworks/models/qwen3p7-plus": {
13000
- id: "accounts/fireworks/models/qwen3p7-plus",
13001
- name: "Qwen 3.7 Plus",
13002
- api: "anthropic-messages",
13003
- provider: "fireworks",
13004
- baseUrl: "https://api.fireworks.ai/inference",
13005
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
13006
- reasoning: true,
13007
- thinkingLevelMap: { "xhigh": "high", "max": "max" },
13008
- input: ["text", "image"],
13009
- cost: {
13010
- input: 0.4,
13011
- output: 1.6,
13012
- cacheRead: 0.08,
13013
- cacheWrite: 0,
13014
- },
13015
- contextWindow: 262144,
13016
- maxTokens: 65536,
13017
- },
13018
13099
  "accounts/fireworks/models/qwen3p8-2p4t-a95b": {
13019
13100
  id: "accounts/fireworks/models/qwen3p8-2p4t-a95b",
13020
13101
  name: "Qwen3.8 2.4T A95B",
@@ -13071,43 +13152,6 @@ export const MODELS = {
13071
13152
  contextWindow: 1000000,
13072
13153
  maxTokens: 384000,
13073
13154
  },
13074
- "accounts/fireworks/routers/deepseek-pro-latest": {
13075
- id: "accounts/fireworks/routers/deepseek-pro-latest",
13076
- name: "DeepSeek Pro Latest",
13077
- api: "anthropic-messages",
13078
- provider: "fireworks",
13079
- baseUrl: "https://api.fireworks.ai/inference",
13080
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
13081
- reasoning: true,
13082
- input: ["text"],
13083
- cost: {
13084
- input: 1.32,
13085
- output: 3.96,
13086
- cacheRead: 0.044,
13087
- cacheWrite: 0,
13088
- },
13089
- contextWindow: 1000000,
13090
- maxTokens: 384000,
13091
- },
13092
- "accounts/fireworks/routers/glm-5p2-fast": {
13093
- id: "accounts/fireworks/routers/glm-5p2-fast",
13094
- name: "GLM 5.2 Fast",
13095
- api: "anthropic-messages",
13096
- provider: "fireworks",
13097
- baseUrl: "https://api.fireworks.ai/inference",
13098
- compat: { "sendSessionAffinityHeaders": true, "supportsEagerToolInputStreaming": false, "supportsCacheControlOnTools": false, "supportsLongCacheRetention": false },
13099
- reasoning: true,
13100
- thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
13101
- input: ["text"],
13102
- cost: {
13103
- input: 2.1,
13104
- output: 6.6,
13105
- cacheRead: 0.21,
13106
- cacheWrite: 0,
13107
- },
13108
- contextWindow: 1048575,
13109
- maxTokens: 131072,
13110
- },
13111
13155
  "accounts/fireworks/routers/glm-5p3-fast": {
13112
13156
  id: "accounts/fireworks/routers/glm-5p3-fast",
13113
13157
  name: "GLM 5.3 Fast",
@@ -13393,6 +13437,26 @@ export const MODELS = {
13393
13437
  contextWindow: 1000000,
13394
13438
  maxTokens: 64000,
13395
13439
  },
13440
+ "claude-opus-5.5": {
13441
+ id: "claude-opus-5.5",
13442
+ name: "Claude Opus 5.5",
13443
+ api: "openai-completions",
13444
+ provider: "github-copilot",
13445
+ baseUrl: "https://api.individual.githubcopilot.com",
13446
+ 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" },
13447
+ compat: { "supportsStore": false, "supportsDeveloperRole": false, "supportsReasoningEffort": false },
13448
+ reasoning: true,
13449
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
13450
+ input: ["text", "image"],
13451
+ cost: {
13452
+ input: 4,
13453
+ output: 20,
13454
+ cacheRead: 0.2,
13455
+ cacheWrite: 5,
13456
+ },
13457
+ contextWindow: 1000000,
13458
+ maxTokens: 128000,
13459
+ },
13396
13460
  "claude-sonnet-4.6": {
13397
13461
  id: "claude-sonnet-4.6",
13398
13462
  name: "Claude Sonnet 4.6",
@@ -13698,6 +13762,63 @@ export const MODELS = {
13698
13762
  contextWindow: 1050000,
13699
13763
  maxTokens: 128000,
13700
13764
  },
13765
+ "gpt-6-luna": {
13766
+ id: "gpt-6-luna",
13767
+ name: "GPT-6 Luna",
13768
+ api: "openai-responses",
13769
+ provider: "github-copilot",
13770
+ baseUrl: "https://api.individual.githubcopilot.com",
13771
+ 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" },
13772
+ reasoning: true,
13773
+ thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
13774
+ input: ["text", "image"],
13775
+ cost: {
13776
+ input: 0.1,
13777
+ output: 0.5,
13778
+ cacheRead: 0.01,
13779
+ cacheWrite: 0.125,
13780
+ },
13781
+ contextWindow: 1050000,
13782
+ maxTokens: 128000,
13783
+ },
13784
+ "gpt-6-sol": {
13785
+ id: "gpt-6-sol",
13786
+ name: "GPT-6 Sol",
13787
+ api: "openai-responses",
13788
+ provider: "github-copilot",
13789
+ baseUrl: "https://api.individual.githubcopilot.com",
13790
+ 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" },
13791
+ reasoning: true,
13792
+ thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
13793
+ input: ["text", "image"],
13794
+ cost: {
13795
+ input: 2,
13796
+ output: 10,
13797
+ cacheRead: 0.2,
13798
+ cacheWrite: 2.5,
13799
+ },
13800
+ contextWindow: 1050000,
13801
+ maxTokens: 128000,
13802
+ },
13803
+ "gpt-6.1-sol": {
13804
+ id: "gpt-6.1-sol",
13805
+ name: "GPT-6.1 Sol",
13806
+ api: "openai-responses",
13807
+ provider: "github-copilot",
13808
+ baseUrl: "https://api.individual.githubcopilot.com",
13809
+ 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" },
13810
+ reasoning: true,
13811
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
13812
+ input: ["text", "image"],
13813
+ cost: {
13814
+ input: 2,
13815
+ output: 10,
13816
+ cacheRead: 0.1,
13817
+ cacheWrite: 2.5,
13818
+ },
13819
+ contextWindow: 1050000,
13820
+ maxTokens: 128000,
13821
+ },
13701
13822
  "grok-4.5": {
13702
13823
  id: "grok-4.5",
13703
13824
  name: "Grok 4.5",
@@ -13883,8 +14004,8 @@ export const MODELS = {
13883
14004
  cacheRead: 0,
13884
14005
  cacheWrite: 0,
13885
14006
  },
13886
- contextWindow: 131072,
13887
- maxTokens: 65536,
14007
+ contextWindow: 128000,
14008
+ maxTokens: 64000,
13888
14009
  },
13889
14010
  "gemini-2.5-flash": {
13890
14011
  id: "gemini-2.5-flash",
@@ -13989,7 +14110,7 @@ export const MODELS = {
13989
14110
  cacheWrite: 0,
13990
14111
  },
13991
14112
  contextWindow: 65536,
13992
- maxTokens: 65536,
14113
+ maxTokens: 4096,
13993
14114
  },
13994
14115
  "gemini-3.1-flash-lite-preview": {
13995
14116
  id: "gemini-3.1-flash-lite-preview",
@@ -16345,7 +16466,7 @@ export const MODELS = {
16345
16466
  cacheRead: 0.06,
16346
16467
  cacheWrite: 0,
16347
16468
  },
16348
- contextWindow: 1048576,
16469
+ contextWindow: 1000000,
16349
16470
  maxTokens: 512000,
16350
16471
  },
16351
16472
  },
@@ -16398,7 +16519,7 @@ export const MODELS = {
16398
16519
  cacheRead: 0.06,
16399
16520
  cacheWrite: 0,
16400
16521
  },
16401
- contextWindow: 1048576,
16522
+ contextWindow: 1000000,
16402
16523
  maxTokens: 512000,
16403
16524
  },
16404
16525
  },
@@ -16414,7 +16535,7 @@ export const MODELS = {
16414
16535
  cost: {
16415
16536
  input: 0.3,
16416
16537
  output: 0.9,
16417
- cacheRead: 0,
16538
+ cacheRead: 0.03,
16418
16539
  cacheWrite: 0,
16419
16540
  },
16420
16541
  contextWindow: 256000,
@@ -16556,23 +16677,6 @@ export const MODELS = {
16556
16677
  contextWindow: 128000,
16557
16678
  maxTokens: 16384,
16558
16679
  },
16559
- "magistral-small": {
16560
- id: "magistral-small",
16561
- name: "Magistral Small",
16562
- api: "mistral-conversations",
16563
- provider: "mistral",
16564
- baseUrl: "https://api.mistral.ai",
16565
- reasoning: true,
16566
- input: ["text"],
16567
- cost: {
16568
- input: 0.5,
16569
- output: 1.5,
16570
- cacheRead: 0,
16571
- cacheWrite: 0,
16572
- },
16573
- contextWindow: 128000,
16574
- maxTokens: 128000,
16575
- },
16576
16680
  "ministral-3b-latest": {
16577
16681
  id: "ministral-3b-latest",
16578
16682
  name: "Ministral 3B (latest)",
@@ -16635,7 +16739,7 @@ export const MODELS = {
16635
16739
  cost: {
16636
16740
  input: 0.5,
16637
16741
  output: 1.5,
16638
- cacheRead: 0,
16742
+ cacheRead: 0.05,
16639
16743
  cacheWrite: 0,
16640
16744
  },
16641
16745
  contextWindow: 262144,
@@ -16652,7 +16756,7 @@ export const MODELS = {
16652
16756
  cost: {
16653
16757
  input: 0.5,
16654
16758
  output: 1.5,
16655
- cacheRead: 0,
16759
+ cacheRead: 0.05,
16656
16760
  cacheWrite: 0,
16657
16761
  },
16658
16762
  contextWindow: 262144,
@@ -16703,7 +16807,7 @@ export const MODELS = {
16703
16807
  cost: {
16704
16808
  input: 1.5,
16705
16809
  output: 7.5,
16706
- cacheRead: 0,
16810
+ cacheRead: 0.15,
16707
16811
  cacheWrite: 0,
16708
16812
  },
16709
16813
  contextWindow: 262144,
@@ -16737,7 +16841,7 @@ export const MODELS = {
16737
16841
  cost: {
16738
16842
  input: 1.5,
16739
16843
  output: 7.5,
16740
- cacheRead: 0,
16844
+ cacheRead: 0.15,
16741
16845
  cacheWrite: 0,
16742
16846
  },
16743
16847
  contextWindow: 262144,
@@ -16788,7 +16892,7 @@ export const MODELS = {
16788
16892
  cost: {
16789
16893
  input: 0.15,
16790
16894
  output: 0.6,
16791
- cacheRead: 0,
16895
+ cacheRead: 0.015,
16792
16896
  cacheWrite: 0,
16793
16897
  },
16794
16898
  contextWindow: 256000,
@@ -16805,7 +16909,7 @@ export const MODELS = {
16805
16909
  cost: {
16806
16910
  input: 0.15,
16807
16911
  output: 0.6,
16808
- cacheRead: 0,
16912
+ cacheRead: 0.015,
16809
16913
  cacheWrite: 0,
16810
16914
  },
16811
16915
  contextWindow: 256000,
@@ -17038,7 +17142,7 @@ export const MODELS = {
17038
17142
  cacheWrite: 0,
17039
17143
  },
17040
17144
  contextWindow: 1048576,
17041
- maxTokens: 131072,
17145
+ maxTokens: 1048576,
17042
17146
  },
17043
17147
  },
17044
17148
  "moonshotai-cn": {
@@ -17112,7 +17216,7 @@ export const MODELS = {
17112
17216
  cacheWrite: 0,
17113
17217
  },
17114
17218
  contextWindow: 1048576,
17115
- maxTokens: 131072,
17219
+ maxTokens: 1048576,
17116
17220
  },
17117
17221
  },
17118
17222
  "nvidia": {
@@ -18101,40 +18205,92 @@ export const MODELS = {
18101
18205
  contextWindow: 1050000,
18102
18206
  maxTokens: 128000,
18103
18207
  },
18104
- "gpt-6-luna": {
18105
- id: "gpt-6-luna",
18106
- name: "GPT-6 Luna",
18208
+ "gpt-6-luna": {
18209
+ id: "gpt-6-luna",
18210
+ name: "GPT-6 Luna",
18211
+ api: "openai-responses",
18212
+ provider: "openai",
18213
+ baseUrl: "https://api.openai.com/v1",
18214
+ reasoning: true,
18215
+ thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
18216
+ input: ["text", "image"],
18217
+ cost: {
18218
+ input: 0.1,
18219
+ output: 0.5,
18220
+ cacheRead: 0.01,
18221
+ cacheWrite: 0.125,
18222
+ },
18223
+ contextWindow: 1050000,
18224
+ maxTokens: 128000,
18225
+ },
18226
+ "gpt-6-sol": {
18227
+ id: "gpt-6-sol",
18228
+ name: "GPT-6 Sol",
18229
+ api: "openai-responses",
18230
+ provider: "openai",
18231
+ baseUrl: "https://api.openai.com/v1",
18232
+ reasoning: true,
18233
+ thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
18234
+ input: ["text", "image"],
18235
+ cost: {
18236
+ input: 2,
18237
+ output: 10,
18238
+ cacheRead: 0.2,
18239
+ cacheWrite: 2.5,
18240
+ },
18241
+ contextWindow: 1050000,
18242
+ maxTokens: 128000,
18243
+ },
18244
+ "gpt-6.1-sol": {
18245
+ id: "gpt-6.1-sol",
18246
+ name: "GPT-6.1 Sol",
18247
+ api: "openai-responses",
18248
+ provider: "openai",
18249
+ baseUrl: "https://api.openai.com/v1",
18250
+ reasoning: true,
18251
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
18252
+ input: ["text", "image"],
18253
+ cost: {
18254
+ input: 2,
18255
+ output: 10,
18256
+ cacheRead: 0.1,
18257
+ cacheWrite: 2.5,
18258
+ },
18259
+ contextWindow: 1050000,
18260
+ maxTokens: 128000,
18261
+ },
18262
+ "gpt-daybreak-blue-latest": {
18263
+ id: "gpt-daybreak-blue-latest",
18264
+ name: "Daybreak Blue",
18107
18265
  api: "openai-responses",
18108
18266
  provider: "openai",
18109
18267
  baseUrl: "https://api.openai.com/v1",
18110
18268
  reasoning: true,
18111
- thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
18112
18269
  input: ["text", "image"],
18113
18270
  cost: {
18114
- input: 0.1,
18115
- output: 0.5,
18116
- cacheRead: 0.01,
18117
- cacheWrite: 0.125,
18271
+ input: 4,
18272
+ output: 20,
18273
+ cacheRead: 0.4,
18274
+ cacheWrite: 5,
18118
18275
  },
18119
18276
  contextWindow: 1050000,
18120
18277
  maxTokens: 128000,
18121
18278
  },
18122
- "gpt-6-sol": {
18123
- id: "gpt-6-sol",
18124
- name: "GPT-6 Sol",
18279
+ "gpt-daybreak-red-latest": {
18280
+ id: "gpt-daybreak-red-latest",
18281
+ name: "Daybreak Red",
18125
18282
  api: "openai-responses",
18126
18283
  provider: "openai",
18127
18284
  baseUrl: "https://api.openai.com/v1",
18128
18285
  reasoning: true,
18129
- thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
18130
18286
  input: ["text", "image"],
18131
18287
  cost: {
18132
- input: 2,
18133
- output: 10,
18134
- cacheRead: 0.2,
18135
- cacheWrite: 2.5,
18288
+ input: 12.5,
18289
+ output: 75,
18290
+ cacheRead: 1.25,
18291
+ cacheWrite: 15.625,
18136
18292
  },
18137
- contextWindow: 1050000,
18293
+ contextWindow: 400000,
18138
18294
  maxTokens: 128000,
18139
18295
  },
18140
18296
  "gpt-realtime-2.1": {
@@ -18402,6 +18558,78 @@ export const MODELS = {
18402
18558
  contextWindow: 1000000,
18403
18559
  maxTokens: 128000,
18404
18560
  },
18561
+ "gpt-6-astra": {
18562
+ id: "gpt-6-astra",
18563
+ name: "GPT-6 Astra",
18564
+ api: "openai-codex-responses",
18565
+ provider: "openai-codex",
18566
+ baseUrl: "https://chatgpt.com/backend-api",
18567
+ reasoning: true,
18568
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": "xhigh" },
18569
+ input: ["text", "image"],
18570
+ cost: {
18571
+ input: 10,
18572
+ output: 50,
18573
+ cacheRead: 1,
18574
+ cacheWrite: 12.5,
18575
+ },
18576
+ contextWindow: 872000,
18577
+ maxTokens: 128000,
18578
+ },
18579
+ "gpt-6-luna": {
18580
+ id: "gpt-6-luna",
18581
+ name: "GPT-6 Luna",
18582
+ api: "openai-codex-responses",
18583
+ provider: "openai-codex",
18584
+ baseUrl: "https://chatgpt.com/backend-api",
18585
+ reasoning: true,
18586
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
18587
+ input: ["text", "image"],
18588
+ cost: {
18589
+ input: 0.1,
18590
+ output: 0.5,
18591
+ cacheRead: 0.01,
18592
+ cacheWrite: 0.125,
18593
+ },
18594
+ contextWindow: 872000,
18595
+ maxTokens: 128000,
18596
+ },
18597
+ "gpt-6-sol": {
18598
+ id: "gpt-6-sol",
18599
+ name: "GPT-6 Sol",
18600
+ api: "openai-codex-responses",
18601
+ provider: "openai-codex",
18602
+ baseUrl: "https://chatgpt.com/backend-api",
18603
+ reasoning: true,
18604
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": "max" },
18605
+ input: ["text", "image"],
18606
+ cost: {
18607
+ input: 2,
18608
+ output: 10,
18609
+ cacheRead: 0.2,
18610
+ cacheWrite: 2.5,
18611
+ },
18612
+ contextWindow: 872000,
18613
+ maxTokens: 128000,
18614
+ },
18615
+ "gpt-6.1-sol": {
18616
+ id: "gpt-6.1-sol",
18617
+ name: "GPT-6.1 Sol",
18618
+ api: "openai-codex-responses",
18619
+ provider: "openai-codex",
18620
+ baseUrl: "https://chatgpt.com/backend-api",
18621
+ reasoning: true,
18622
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": "xhigh" },
18623
+ input: ["text", "image"],
18624
+ cost: {
18625
+ input: 2,
18626
+ output: 10,
18627
+ cacheRead: 0.1,
18628
+ cacheWrite: 2.5,
18629
+ },
18630
+ contextWindow: 872000,
18631
+ maxTokens: 128000,
18632
+ },
18405
18633
  },
18406
18634
  "opencode": {
18407
18635
  "big-pickle": {
@@ -19345,6 +19573,24 @@ export const MODELS = {
19345
19573
  contextWindow: 1050000,
19346
19574
  maxTokens: 128000,
19347
19575
  },
19576
+ "gpt-6.1-sol": {
19577
+ id: "gpt-6.1-sol",
19578
+ name: "GPT-6.1 Sol",
19579
+ api: "openai-responses",
19580
+ provider: "opencode",
19581
+ baseUrl: "https://opencode.ai/zen/v1",
19582
+ reasoning: true,
19583
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
19584
+ input: ["text", "image"],
19585
+ cost: {
19586
+ input: 2,
19587
+ output: 10,
19588
+ cacheRead: 0.1,
19589
+ cacheWrite: 2.5,
19590
+ },
19591
+ contextWindow: 1050000,
19592
+ maxTokens: 128000,
19593
+ },
19348
19594
  "grok-4.5": {
19349
19595
  id: "grok-4.5",
19350
19596
  name: "Grok 4.5",
@@ -19380,6 +19626,24 @@ export const MODELS = {
19380
19626
  contextWindow: 500000,
19381
19627
  maxTokens: 500000,
19382
19628
  },
19629
+ "grok-4.7": {
19630
+ id: "grok-4.7",
19631
+ name: "Grok 4.7",
19632
+ api: "openai-responses",
19633
+ provider: "opencode",
19634
+ baseUrl: "https://opencode.ai/zen/v1",
19635
+ reasoning: true,
19636
+ thinkingLevelMap: { "xhigh": "xhigh" },
19637
+ input: ["text", "image"],
19638
+ cost: {
19639
+ input: 2,
19640
+ output: 6,
19641
+ cacheRead: 0.5,
19642
+ cacheWrite: 0,
19643
+ },
19644
+ contextWindow: 500000,
19645
+ maxTokens: 500000,
19646
+ },
19383
19647
  "grok-build-0.1": {
19384
19648
  id: "grok-build-0.1",
19385
19649
  name: "Grok Build 0.1",
@@ -19484,6 +19748,23 @@ export const MODELS = {
19484
19748
  contextWindow: 262144,
19485
19749
  maxTokens: 32768,
19486
19750
  },
19751
+ "longcat-2.5-preview-free": {
19752
+ id: "longcat-2.5-preview-free",
19753
+ name: "LongCat 2.5 Preview Free",
19754
+ api: "openai-completions",
19755
+ provider: "opencode",
19756
+ baseUrl: "https://opencode.ai/zen/v1",
19757
+ reasoning: true,
19758
+ input: ["text", "image"],
19759
+ cost: {
19760
+ input: 0,
19761
+ output: 0,
19762
+ cacheRead: 0,
19763
+ cacheWrite: 0,
19764
+ },
19765
+ contextWindow: 1000000,
19766
+ maxTokens: 131072,
19767
+ },
19487
19768
  "mimo-v2.6-flash-free": {
19488
19769
  id: "mimo-v2.6-flash-free",
19489
19770
  name: "MiMo-V2.6-Flash Free",
@@ -19570,24 +19851,6 @@ export const MODELS = {
19570
19851
  contextWindow: 1048576,
19571
19852
  maxTokens: 131072,
19572
19853
  },
19573
- "muse-spark-1.2-contributor-free": {
19574
- id: "muse-spark-1.2-contributor-free",
19575
- name: "Muse Spark 1.2 Free",
19576
- api: "openai-responses",
19577
- provider: "opencode",
19578
- baseUrl: "https://opencode.ai/zen/v1",
19579
- reasoning: true,
19580
- thinkingLevelMap: { "off": null, "minimal": "minimal", "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "xhigh" },
19581
- input: ["text", "image"],
19582
- cost: {
19583
- input: 0,
19584
- output: 0,
19585
- cacheRead: 0,
19586
- cacheWrite: 0,
19587
- },
19588
- contextWindow: 1048576,
19589
- maxTokens: 131072,
19590
- },
19591
19854
  "muse-spark-1.3": {
19592
19855
  id: "muse-spark-1.3",
19593
19856
  name: "Muse Spark 1.3",
@@ -19712,6 +19975,41 @@ export const MODELS = {
19712
19975
  contextWindow: 1000000,
19713
19976
  maxTokens: 131072,
19714
19977
  },
19978
+ "qwen3.8-max": {
19979
+ id: "qwen3.8-max",
19980
+ name: "Qwen3.8 Max",
19981
+ api: "openai-completions",
19982
+ provider: "opencode",
19983
+ baseUrl: "https://opencode.ai/zen/v1",
19984
+ reasoning: true,
19985
+ thinkingLevelMap: { "xhigh": "high", "max": "max" },
19986
+ input: ["text", "image"],
19987
+ cost: {
19988
+ input: 2,
19989
+ output: 6,
19990
+ cacheRead: 0.25,
19991
+ cacheWrite: 2.5,
19992
+ },
19993
+ contextWindow: 262144,
19994
+ maxTokens: 131072,
19995
+ },
19996
+ "space-bunny-free": {
19997
+ id: "space-bunny-free",
19998
+ name: "Space Bunny Free",
19999
+ api: "openai-completions",
20000
+ provider: "opencode",
20001
+ baseUrl: "https://opencode.ai/zen/v1",
20002
+ reasoning: true,
20003
+ input: ["text", "image"],
20004
+ cost: {
20005
+ input: 0,
20006
+ output: 0,
20007
+ cacheRead: 0,
20008
+ cacheWrite: 0,
20009
+ },
20010
+ contextWindow: 1048576,
20011
+ maxTokens: 524288,
20012
+ },
19715
20013
  },
19716
20014
  "opencode-go": {
19717
20015
  "deepseek-v4-flash": {
@@ -19790,23 +20088,6 @@ export const MODELS = {
19790
20088
  contextWindow: 1000000,
19791
20089
  maxTokens: 384000,
19792
20090
  },
19793
- "glm-5.1": {
19794
- id: "glm-5.1",
19795
- name: "GLM-5.1",
19796
- api: "openai-completions",
19797
- provider: "opencode-go",
19798
- baseUrl: "https://opencode.ai/zen/go/v1",
19799
- reasoning: true,
19800
- input: ["text"],
19801
- cost: {
19802
- input: 1.4,
19803
- output: 4.4,
19804
- cacheRead: 0.26,
19805
- cacheWrite: 0,
19806
- },
19807
- contextWindow: 202752,
19808
- maxTokens: 32768,
19809
- },
19810
20091
  "glm-5.2": {
19811
20092
  id: "glm-5.2",
19812
20093
  name: "GLM-5.2",
@@ -19879,6 +20160,24 @@ export const MODELS = {
19879
20160
  contextWindow: 1000000,
19880
20161
  maxTokens: 128000,
19881
20162
  },
20163
+ "gpt-6-luna": {
20164
+ id: "gpt-6-luna",
20165
+ name: "GPT-6 Luna",
20166
+ api: "openai-responses",
20167
+ provider: "opencode-go",
20168
+ baseUrl: "https://opencode.ai/zen/go/v1",
20169
+ reasoning: true,
20170
+ thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
20171
+ input: ["text", "image"],
20172
+ cost: {
20173
+ input: 0.1,
20174
+ output: 0.5,
20175
+ cacheRead: 0.01,
20176
+ cacheWrite: 0.125,
20177
+ },
20178
+ contextWindow: 1050000,
20179
+ maxTokens: 128000,
20180
+ },
19882
20181
  "grok-4.6": {
19883
20182
  id: "grok-4.6",
19884
20183
  name: "Grok 4.6",
@@ -19949,25 +20248,6 @@ export const MODELS = {
19949
20248
  contextWindow: 1024000,
19950
20249
  maxTokens: 64000,
19951
20250
  },
19952
- "kimi-k2.6": {
19953
- id: "kimi-k2.6",
19954
- name: "Kimi K2.6",
19955
- api: "openai-completions",
19956
- provider: "opencode-go",
19957
- baseUrl: "https://opencode.ai/zen/go/v1",
19958
- compat: { "thinkingFormat": "deepseek", "supportsReasoningEffort": false },
19959
- reasoning: true,
19960
- thinkingLevelMap: { "minimal": null, "low": null, "medium": null },
19961
- input: ["text", "image"],
19962
- cost: {
19963
- input: 0.95,
19964
- output: 4,
19965
- cacheRead: 0.16,
19966
- cacheWrite: 0,
19967
- },
19968
- contextWindow: 262144,
19969
- maxTokens: 65536,
19970
- },
19971
20251
  "kimi-k2.7-code": {
19972
20252
  id: "kimi-k2.7-code",
19973
20253
  name: "Kimi K2.7 Code",
@@ -19999,21 +20279,38 @@ export const MODELS = {
19999
20279
  cacheRead: 0.3,
20000
20280
  cacheWrite: 0,
20001
20281
  },
20002
- contextWindow: 1048576,
20282
+ contextWindow: 1048576,
20283
+ maxTokens: 131072,
20284
+ },
20285
+ "longcat-2.0": {
20286
+ id: "longcat-2.0",
20287
+ name: "LongCat-2.0",
20288
+ api: "openai-completions",
20289
+ provider: "opencode-go",
20290
+ baseUrl: "https://opencode.ai/zen/go/v1",
20291
+ reasoning: true,
20292
+ input: ["text"],
20293
+ cost: {
20294
+ input: 0.3,
20295
+ output: 1.2,
20296
+ cacheRead: 0.006,
20297
+ cacheWrite: 0,
20298
+ },
20299
+ contextWindow: 1000000,
20003
20300
  maxTokens: 131072,
20004
20301
  },
20005
- "longcat-2.0": {
20006
- id: "longcat-2.0",
20007
- name: "LongCat-2.0",
20302
+ "longcat-2.5-preview-free": {
20303
+ id: "longcat-2.5-preview-free",
20304
+ name: "LongCat 2.5 Preview Free",
20008
20305
  api: "openai-completions",
20009
20306
  provider: "opencode-go",
20010
20307
  baseUrl: "https://opencode.ai/zen/go/v1",
20011
20308
  reasoning: true,
20012
- input: ["text"],
20309
+ input: ["text", "image"],
20013
20310
  cost: {
20014
- input: 0.3,
20015
- output: 1.2,
20016
- cacheRead: 0.006,
20311
+ input: 0,
20312
+ output: 0,
20313
+ cacheRead: 0,
20017
20314
  cacheWrite: 0,
20018
20315
  },
20019
20316
  contextWindow: 1000000,
@@ -20157,43 +20454,6 @@ export const MODELS = {
20157
20454
  contextWindow: 1048576,
20158
20455
  maxTokens: 131072,
20159
20456
  },
20160
- "qwen3.6-plus": {
20161
- id: "qwen3.6-plus",
20162
- name: "Qwen3.6 Plus",
20163
- api: "openai-completions",
20164
- provider: "opencode-go",
20165
- baseUrl: "https://opencode.ai/zen/go/v1",
20166
- compat: { "thinkingFormat": "qwen" },
20167
- reasoning: true,
20168
- thinkingLevelMap: { "xhigh": "high", "max": "max" },
20169
- input: ["text", "image"],
20170
- cost: {
20171
- input: 0.5,
20172
- output: 3,
20173
- cacheRead: 0.05,
20174
- cacheWrite: 0.625,
20175
- },
20176
- contextWindow: 1000000,
20177
- maxTokens: 65536,
20178
- },
20179
- "qwen3.7-max": {
20180
- id: "qwen3.7-max",
20181
- name: "Qwen3.7 Max",
20182
- api: "openai-completions",
20183
- provider: "opencode-go",
20184
- baseUrl: "https://opencode.ai/zen/go/v1",
20185
- reasoning: true,
20186
- thinkingLevelMap: { "xhigh": "high", "max": "max" },
20187
- input: ["text"],
20188
- cost: {
20189
- input: 2.5,
20190
- output: 7.5,
20191
- cacheRead: 0.5,
20192
- cacheWrite: 3.125,
20193
- },
20194
- contextWindow: 1000000,
20195
- maxTokens: 65536,
20196
- },
20197
20457
  "qwen3.7-plus": {
20198
20458
  id: "qwen3.7-plus",
20199
20459
  name: "Qwen3.7 Plus",
@@ -20248,6 +20508,23 @@ export const MODELS = {
20248
20508
  contextWindow: 1000000,
20249
20509
  maxTokens: 131072,
20250
20510
  },
20511
+ "space-bunny-free": {
20512
+ id: "space-bunny-free",
20513
+ name: "Space Bunny Free",
20514
+ api: "openai-completions",
20515
+ provider: "opencode-go",
20516
+ baseUrl: "https://opencode.ai/zen/go/v1",
20517
+ reasoning: true,
20518
+ input: ["text", "image"],
20519
+ cost: {
20520
+ input: 0,
20521
+ output: 0,
20522
+ cacheRead: 0,
20523
+ cacheWrite: 0,
20524
+ },
20525
+ contextWindow: 1048576,
20526
+ maxTokens: 524288,
20527
+ },
20251
20528
  },
20252
20529
  "openrouter": {
20253
20530
  "aion-labs/aion-2.0": {
@@ -20265,7 +20542,7 @@ export const MODELS = {
20265
20542
  cacheRead: 0.19999999999999998,
20266
20543
  cacheWrite: 0,
20267
20544
  },
20268
- contextWindow: 1048576,
20545
+ contextWindow: 131072,
20269
20546
  maxTokens: 32768,
20270
20547
  },
20271
20548
  "aion-labs/aion-3.0": {
@@ -20283,7 +20560,7 @@ export const MODELS = {
20283
20560
  cacheRead: 0.75,
20284
20561
  cacheWrite: 0,
20285
20562
  },
20286
- contextWindow: 1048576,
20563
+ contextWindow: 131072,
20287
20564
  maxTokens: 32768,
20288
20565
  },
20289
20566
  "aion-labs/aion-3.0-mini": {
@@ -20301,7 +20578,43 @@ export const MODELS = {
20301
20578
  cacheRead: 0.18,
20302
20579
  cacheWrite: 0,
20303
20580
  },
20304
- contextWindow: 1048576,
20581
+ contextWindow: 131072,
20582
+ maxTokens: 32768,
20583
+ },
20584
+ "aion-labs/aion-3.5": {
20585
+ id: "aion-labs/aion-3.5",
20586
+ name: "AionLabs: Aion 3.5",
20587
+ api: "openai-completions",
20588
+ provider: "openrouter",
20589
+ baseUrl: "https://openrouter.ai/api/v1",
20590
+ reasoning: true,
20591
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
20592
+ input: ["text"],
20593
+ cost: {
20594
+ input: 3,
20595
+ output: 6,
20596
+ cacheRead: 0.75,
20597
+ cacheWrite: 0,
20598
+ },
20599
+ contextWindow: 262144,
20600
+ maxTokens: 32768,
20601
+ },
20602
+ "aion-labs/aion-3.5-mini": {
20603
+ id: "aion-labs/aion-3.5-mini",
20604
+ name: "AionLabs: Aion 3.5 Mini",
20605
+ api: "openai-completions",
20606
+ provider: "openrouter",
20607
+ baseUrl: "https://openrouter.ai/api/v1",
20608
+ reasoning: true,
20609
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
20610
+ input: ["text"],
20611
+ cost: {
20612
+ input: 0.7,
20613
+ output: 1.4,
20614
+ cacheRead: 0.18,
20615
+ cacheWrite: 0,
20616
+ },
20617
+ contextWindow: 262144,
20305
20618
  maxTokens: 32768,
20306
20619
  },
20307
20620
  "amazon/nova-2-lite-v1": {
@@ -20389,23 +20702,6 @@ export const MODELS = {
20389
20702
  contextWindow: 300000,
20390
20703
  maxTokens: 5120,
20391
20704
  },
20392
- "anthropic/claude-3-haiku": {
20393
- id: "anthropic/claude-3-haiku",
20394
- name: "Anthropic: Claude 3 Haiku",
20395
- api: "openai-completions",
20396
- provider: "openrouter",
20397
- baseUrl: "https://openrouter.ai/api/v1",
20398
- reasoning: false,
20399
- input: ["text", "image"],
20400
- cost: {
20401
- input: 0.25,
20402
- output: 1.25,
20403
- cacheRead: 0.03,
20404
- cacheWrite: 0.3,
20405
- },
20406
- contextWindow: 200000,
20407
- maxTokens: 4096,
20408
- },
20409
20705
  "anthropic/claude-fable-5": {
20410
20706
  id: "anthropic/claude-fable-5",
20411
20707
  name: "Anthropic: Claude Fable 5",
@@ -21023,6 +21319,23 @@ export const MODELS = {
21023
21319
  contextWindow: 262144,
21024
21320
  maxTokens: 131072,
21025
21321
  },
21322
+ "cohere/command-a-plus": {
21323
+ id: "cohere/command-a-plus",
21324
+ name: "Cohere: Command A+",
21325
+ api: "openai-completions",
21326
+ provider: "openrouter",
21327
+ baseUrl: "https://openrouter.ai/api/v1",
21328
+ reasoning: true,
21329
+ input: ["text", "image"],
21330
+ cost: {
21331
+ input: 0.3,
21332
+ output: 1.5,
21333
+ cacheRead: 0.15,
21334
+ cacheWrite: 0,
21335
+ },
21336
+ contextWindow: 192000,
21337
+ maxTokens: 64000,
21338
+ },
21026
21339
  "cohere/command-r-08-2024": {
21027
21340
  id: "cohere/command-r-08-2024",
21028
21341
  name: "Cohere: Command R (08-2024)",
@@ -21083,13 +21396,13 @@ export const MODELS = {
21083
21396
  reasoning: false,
21084
21397
  input: ["text"],
21085
21398
  cost: {
21086
- input: 0.32,
21087
- output: 0.8899999999999999,
21399
+ input: 0.2574,
21400
+ output: 1.0287,
21088
21401
  cacheRead: 0,
21089
21402
  cacheWrite: 0,
21090
21403
  },
21091
21404
  contextWindow: 163840,
21092
- maxTokens: 16384,
21405
+ maxTokens: 16000,
21093
21406
  },
21094
21407
  "deepseek/deepseek-chat-v3-0324": {
21095
21408
  id: "deepseek/deepseek-chat-v3-0324",
@@ -21100,9 +21413,9 @@ export const MODELS = {
21100
21413
  reasoning: false,
21101
21414
  input: ["text"],
21102
21415
  cost: {
21103
- input: 0.25,
21104
- output: 1,
21105
- cacheRead: 0,
21416
+ input: 0.29,
21417
+ output: 1.1400000000000001,
21418
+ cacheRead: 0.11,
21106
21419
  cacheWrite: 0,
21107
21420
  },
21108
21421
  contextWindow: 163840,
@@ -21170,13 +21483,13 @@ export const MODELS = {
21170
21483
  reasoning: true,
21171
21484
  input: ["text"],
21172
21485
  cost: {
21173
- input: 0.27,
21486
+ input: 0.3,
21174
21487
  output: 1,
21175
21488
  cacheRead: 0.135,
21176
21489
  cacheWrite: 0,
21177
21490
  },
21178
21491
  contextWindow: 163840,
21179
- maxTokens: 32768,
21492
+ maxTokens: 65536,
21180
21493
  },
21181
21494
  "deepseek/deepseek-v3.2": {
21182
21495
  id: "deepseek/deepseek-v3.2",
@@ -21187,9 +21500,9 @@ export const MODELS = {
21187
21500
  reasoning: true,
21188
21501
  input: ["text"],
21189
21502
  cost: {
21190
- input: 0.26899999999999996,
21191
- output: 0.39999999999999997,
21192
- cacheRead: 0.13449999999999998,
21503
+ input: 0.28,
21504
+ output: 0.42,
21505
+ cacheRead: 0.028,
21193
21506
  cacheWrite: 0,
21194
21507
  },
21195
21508
  contextWindow: 163840,
@@ -21210,7 +21523,7 @@ export const MODELS = {
21210
21523
  cacheWrite: 0,
21211
21524
  },
21212
21525
  contextWindow: 163840,
21213
- maxTokens: 65536,
21526
+ maxTokens: 147456,
21214
21527
  },
21215
21528
  "deepseek/deepseek-v4-flash": {
21216
21529
  id: "deepseek/deepseek-v4-flash",
@@ -21223,13 +21536,13 @@ export const MODELS = {
21223
21536
  thinkingLevelMap: { "minimal": null, "low": null, "medium": null, "high": "high", "xhigh": "xhigh", "off": "none", "max": null, "ultra": null },
21224
21537
  input: ["text"],
21225
21538
  cost: {
21226
- input: 0.049,
21227
- output: 0.098,
21228
- cacheRead: 0.0098,
21539
+ input: 0.14,
21540
+ output: 0.28,
21541
+ cacheRead: 0.028,
21229
21542
  cacheWrite: 0,
21230
21543
  },
21231
21544
  contextWindow: 1048576,
21232
- maxTokens: 384000,
21545
+ maxTokens: 131072,
21233
21546
  },
21234
21547
  "deepseek/deepseek-v4-flash-0731": {
21235
21548
  id: "deepseek/deepseek-v4-flash-0731",
@@ -21242,12 +21555,12 @@ export const MODELS = {
21242
21555
  thinkingLevelMap: { "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "off": "none", "max": "max", "ultra": null },
21243
21556
  input: ["text"],
21244
21557
  cost: {
21245
- input: 0.04,
21246
- output: 0.64,
21247
- cacheRead: 0.016,
21558
+ input: 0.018,
21559
+ output: 0.32,
21560
+ cacheRead: 0.018,
21248
21561
  cacheWrite: 0,
21249
21562
  },
21250
- contextWindow: 1310720,
21563
+ contextWindow: 1048576,
21251
21564
  maxTokens: 943718,
21252
21565
  },
21253
21566
  "deepseek/deepseek-v4-flash-vision-exp": {
@@ -21261,13 +21574,13 @@ export const MODELS = {
21261
21574
  thinkingLevelMap: { "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "off": "none", "max": "max", "ultra": null },
21262
21575
  input: ["text", "image"],
21263
21576
  cost: {
21264
- input: 0.22,
21265
- output: 0.66,
21266
- cacheRead: 0.007,
21577
+ input: 0.21559999999999999,
21578
+ output: 0.6468,
21579
+ cacheRead: 0.00686,
21267
21580
  cacheWrite: 0,
21268
21581
  },
21269
21582
  contextWindow: 1048576,
21270
- maxTokens: 943718,
21583
+ maxTokens: 262144,
21271
21584
  },
21272
21585
  "deepseek/deepseek-v4-pro": {
21273
21586
  id: "deepseek/deepseek-v4-pro",
@@ -21280,9 +21593,9 @@ export const MODELS = {
21280
21593
  thinkingLevelMap: { "minimal": null, "low": null, "medium": null, "high": "high", "xhigh": "xhigh", "off": "none", "max": null, "ultra": null },
21281
21594
  input: ["text"],
21282
21595
  cost: {
21283
- input: 0.895578,
21284
- output: 1.791156,
21285
- cacheRead: 0.0746315,
21596
+ input: 0.9552599999999999,
21597
+ output: 1.9105199999999998,
21598
+ cacheRead: 0.07960500000000001,
21286
21599
  cacheWrite: 0,
21287
21600
  },
21288
21601
  contextWindow: 1048576,
@@ -21299,13 +21612,13 @@ export const MODELS = {
21299
21612
  thinkingLevelMap: { "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "off": "none", "max": "max", "ultra": null },
21300
21613
  input: ["text"],
21301
21614
  cost: {
21302
- input: 0.66,
21303
- output: 1.9800000000000002,
21304
- cacheRead: 0.022,
21615
+ input: 1.32,
21616
+ output: 3.9600000000000004,
21617
+ cacheRead: 0.044,
21305
21618
  cacheWrite: 0,
21306
21619
  },
21307
21620
  contextWindow: 1048576,
21308
- maxTokens: 384000,
21621
+ maxTokens: 393216,
21309
21622
  },
21310
21623
  "deepseek/deepseek-v4.1-flash": {
21311
21624
  id: "deepseek/deepseek-v4.1-flash",
@@ -21318,13 +21631,13 @@ export const MODELS = {
21318
21631
  thinkingLevelMap: { "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "off": "none", "max": "max", "ultra": null },
21319
21632
  input: ["text", "image"],
21320
21633
  cost: {
21321
- input: 0.15,
21322
- output: 0.6,
21323
- cacheRead: 0.003,
21634
+ input: 0.3,
21635
+ output: 1.2,
21636
+ cacheRead: 0.006,
21324
21637
  cacheWrite: 0,
21325
21638
  },
21326
21639
  contextWindow: 1048576,
21327
- maxTokens: 384000,
21640
+ maxTokens: 943718,
21328
21641
  },
21329
21642
  "deepseek/deepseek-v4.1-flash:batch": {
21330
21643
  id: "deepseek/deepseek-v4.1-flash:batch",
@@ -21362,6 +21675,24 @@ export const MODELS = {
21362
21675
  contextWindow: 512000,
21363
21676
  maxTokens: 460800,
21364
21677
  },
21678
+ "fireworks/ember-1": {
21679
+ id: "fireworks/ember-1",
21680
+ name: "Fireworks: Ember-1",
21681
+ api: "openai-completions",
21682
+ provider: "openrouter",
21683
+ baseUrl: "https://openrouter.ai/api/v1",
21684
+ reasoning: true,
21685
+ thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
21686
+ input: ["text", "image"],
21687
+ cost: {
21688
+ input: 3,
21689
+ output: 15,
21690
+ cacheRead: 0.3,
21691
+ cacheWrite: 0,
21692
+ },
21693
+ contextWindow: 1048576,
21694
+ maxTokens: 943718,
21695
+ },
21365
21696
  "google/gemini-2.5-flash": {
21366
21697
  id: "google/gemini-2.5-flash",
21367
21698
  name: "Google: Gemini 2.5 Flash",
@@ -21869,9 +22200,9 @@ export const MODELS = {
21869
22200
  reasoning: true,
21870
22201
  input: ["text", "image"],
21871
22202
  cost: {
21872
- input: 0.09,
21873
- output: 0.3,
21874
- cacheRead: 0.049999999999999996,
22203
+ input: 0.0765,
22204
+ output: 0.255,
22205
+ cacheRead: 0.042499999999999996,
21875
22206
  cacheWrite: 0,
21876
22207
  },
21877
22208
  contextWindow: 262144,
@@ -22016,23 +22347,6 @@ export const MODELS = {
22016
22347
  contextWindow: 262144,
22017
22348
  maxTokens: 235929,
22018
22349
  },
22019
- "inclusionai/ling-3.0-flash-fin:free": {
22020
- id: "inclusionai/ling-3.0-flash-fin:free",
22021
- name: "inclusionAI: Ling 3.0 Flash Fin (free)",
22022
- api: "openai-completions",
22023
- provider: "openrouter",
22024
- baseUrl: "https://openrouter.ai/api/v1",
22025
- reasoning: true,
22026
- input: ["text"],
22027
- cost: {
22028
- input: 0,
22029
- output: 0,
22030
- cacheRead: 0,
22031
- cacheWrite: 0,
22032
- },
22033
- contextWindow: 262144,
22034
- maxTokens: 32768,
22035
- },
22036
22350
  "inclusionai/ling-3.0-flash-sante:free": {
22037
22351
  id: "inclusionai/ling-3.0-flash-sante:free",
22038
22352
  name: "inclusionAI: Ling 3.0 Flash Sante (free)",
@@ -22059,26 +22373,9 @@ export const MODELS = {
22059
22373
  reasoning: true,
22060
22374
  input: ["text", "image"],
22061
22375
  cost: {
22062
- input: 0.06,
22063
- output: 0.18,
22064
- cacheRead: 0.012,
22065
- cacheWrite: 0,
22066
- },
22067
- contextWindow: 131072,
22068
- maxTokens: 32768,
22069
- },
22070
- "inclusionai/ling-3.0-flash-vl:free": {
22071
- id: "inclusionai/ling-3.0-flash-vl:free",
22072
- name: "inclusionAI: Ling 3.0 Flash VL (free)",
22073
- api: "openai-completions",
22074
- provider: "openrouter",
22075
- baseUrl: "https://openrouter.ai/api/v1",
22076
- reasoning: true,
22077
- input: ["text", "image"],
22078
- cost: {
22079
- input: 0,
22080
- output: 0,
22081
- cacheRead: 0,
22376
+ input: 0.020999999999999998,
22377
+ output: 0.061599999999999995,
22378
+ cacheRead: 0.004200000000000001,
22082
22379
  cacheWrite: 0,
22083
22380
  },
22084
22381
  contextWindow: 262144,
@@ -22231,13 +22528,13 @@ export const MODELS = {
22231
22528
  thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": null, "ultra": null },
22232
22529
  input: ["text", "image"],
22233
22530
  cost: {
22234
- input: 0.3,
22235
- output: 1.2,
22531
+ input: 0.35,
22532
+ output: 1.5,
22236
22533
  cacheRead: 0.04,
22237
22534
  cacheWrite: 0,
22238
22535
  },
22239
22536
  contextWindow: 131072,
22240
- maxTokens: 16384,
22537
+ maxTokens: 117964,
22241
22538
  },
22242
22539
  "meta/muse-spark-1.1": {
22243
22540
  id: "meta/muse-spark-1.1",
@@ -22356,13 +22653,13 @@ export const MODELS = {
22356
22653
  thinkingLevelMap: { "off": null },
22357
22654
  input: ["text"],
22358
22655
  cost: {
22359
- input: 0.255,
22360
- output: 1.02,
22656
+ input: 0.3,
22657
+ output: 1.2,
22361
22658
  cacheRead: 0,
22362
22659
  cacheWrite: 0,
22363
22660
  },
22364
22661
  contextWindow: 204800,
22365
- maxTokens: 131072,
22662
+ maxTokens: 176947,
22366
22663
  },
22367
22664
  "minimax/minimax-m2.1": {
22368
22665
  id: "minimax/minimax-m2.1",
@@ -22410,13 +22707,13 @@ export const MODELS = {
22410
22707
  thinkingLevelMap: { "off": null },
22411
22708
  input: ["text"],
22412
22709
  cost: {
22413
- input: 0.3,
22414
- output: 1.2,
22415
- cacheRead: 0.06,
22710
+ input: 0.21,
22711
+ output: 0.84,
22712
+ cacheRead: 0.041999999999999996,
22416
22713
  cacheWrite: 0,
22417
22714
  },
22418
22715
  contextWindow: 204800,
22419
- maxTokens: 131072,
22716
+ maxTokens: 176947,
22420
22717
  },
22421
22718
  "minimax/minimax-m3": {
22422
22719
  id: "minimax/minimax-m3",
@@ -22585,8 +22882,25 @@ export const MODELS = {
22585
22882
  cacheRead: 0.19999999999999998,
22586
22883
  cacheWrite: 0,
22587
22884
  },
22588
- contextWindow: 131072,
22589
- maxTokens: 104857,
22885
+ contextWindow: 131072,
22886
+ maxTokens: 104857,
22887
+ },
22888
+ "mistralai/mistral-large-2512": {
22889
+ id: "mistralai/mistral-large-2512",
22890
+ name: "Mistral: Mistral Large 3 2512",
22891
+ api: "openai-completions",
22892
+ provider: "openrouter",
22893
+ baseUrl: "https://openrouter.ai/api/v1",
22894
+ reasoning: false,
22895
+ input: ["text", "image"],
22896
+ cost: {
22897
+ input: 0.5,
22898
+ output: 1.5,
22899
+ cacheRead: 0.049999999999999996,
22900
+ cacheWrite: 0,
22901
+ },
22902
+ contextWindow: 262144,
22903
+ maxTokens: 209715,
22590
22904
  },
22591
22905
  "mistralai/mistral-large-2512:batch": {
22592
22906
  id: "mistralai/mistral-large-2512:batch",
@@ -22909,9 +23223,9 @@ export const MODELS = {
22909
23223
  reasoning: true,
22910
23224
  input: ["text", "image"],
22911
23225
  cost: {
22912
- input: 0.95,
22913
- output: 4,
22914
- cacheRead: 0.16,
23226
+ input: 0.65,
23227
+ output: 3.41,
23228
+ cacheRead: 0.15,
22915
23229
  cacheWrite: 0,
22916
23230
  },
22917
23231
  contextWindow: 262144,
@@ -22927,7 +23241,7 @@ export const MODELS = {
22927
23241
  thinkingLevelMap: { "off": null },
22928
23242
  input: ["text", "image"],
22929
23243
  cost: {
22930
- input: 0.7062,
23244
+ input: 0.6562,
22931
23245
  output: 3.3000000000000003,
22932
23246
  cacheRead: 0.18,
22933
23247
  cacheWrite: 0,
@@ -22971,24 +23285,6 @@ export const MODELS = {
22971
23285
  contextWindow: 1048576,
22972
23286
  maxTokens: 16384,
22973
23287
  },
22974
- "nex-agi/nex-n2.5-mini:free": {
22975
- id: "nex-agi/nex-n2.5-mini:free",
22976
- name: "Nex AGI: Nex-N2.5-Mini (free)",
22977
- api: "openai-completions",
22978
- provider: "openrouter",
22979
- baseUrl: "https://openrouter.ai/api/v1",
22980
- reasoning: true,
22981
- thinkingLevelMap: { "off": "none", "minimal": null, "low": null, "medium": "medium", "high": "high", "xhigh": null, "max": null, "ultra": null },
22982
- input: ["text", "image"],
22983
- cost: {
22984
- input: 0,
22985
- output: 0,
22986
- cacheRead: 0,
22987
- cacheWrite: 0,
22988
- },
22989
- contextWindow: 262144,
22990
- maxTokens: 235929,
22991
- },
22992
23288
  "nex-agi/nex-n2.5-pro": {
22993
23289
  id: "nex-agi/nex-n2.5-pro",
22994
23290
  name: "Nex AGI: Nex-N2.5-Pro",
@@ -23007,24 +23303,6 @@ export const MODELS = {
23007
23303
  contextWindow: 262144,
23008
23304
  maxTokens: 235929,
23009
23305
  },
23010
- "nex-agi/nex-n2.5-pro:free": {
23011
- id: "nex-agi/nex-n2.5-pro:free",
23012
- name: "Nex AGI: Nex-N2.5-Pro (free)",
23013
- api: "openai-completions",
23014
- provider: "openrouter",
23015
- baseUrl: "https://openrouter.ai/api/v1",
23016
- reasoning: true,
23017
- thinkingLevelMap: { "off": "none", "minimal": null, "low": null, "medium": "medium", "high": "high", "xhigh": null, "max": null, "ultra": null },
23018
- input: ["text", "image"],
23019
- cost: {
23020
- input: 0,
23021
- output: 0,
23022
- cacheRead: 0,
23023
- cacheWrite: 0,
23024
- },
23025
- contextWindow: 262144,
23026
- maxTokens: 235929,
23027
- },
23028
23306
  "nvidia/nemotron-3-nano-30b-a3b": {
23029
23307
  id: "nvidia/nemotron-3-nano-30b-a3b",
23030
23308
  name: "NVIDIA: Nemotron 3 Nano 30B A3B",
@@ -23140,13 +23418,13 @@ export const MODELS = {
23140
23418
  reasoning: true,
23141
23419
  input: ["text"],
23142
23420
  cost: {
23143
- input: 0.07,
23144
- output: 0.19999999999999998,
23145
- cacheRead: 0.04,
23421
+ input: 0.06,
23422
+ output: 0.16,
23423
+ cacheRead: 0.03,
23146
23424
  cacheWrite: 0,
23147
23425
  },
23148
23426
  contextWindow: 262144,
23149
- maxTokens: 235929,
23427
+ maxTokens: 32768,
23150
23428
  },
23151
23429
  "nvidia/nemotron-3.5-lightning:free": {
23152
23430
  id: "nvidia/nemotron-3.5-lightning:free",
@@ -24180,10 +24458,10 @@ export const MODELS = {
24180
24458
  thinkingLevelMap: { "xhigh": "xhigh", "max": "max", "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "ultra": null },
24181
24459
  input: ["text", "image"],
24182
24460
  cost: {
24183
- input: 2,
24184
- output: 10,
24185
- cacheRead: 0.19999999999999998,
24186
- cacheWrite: 2.5,
24461
+ input: 4,
24462
+ output: 20,
24463
+ cacheRead: 0.39999999999999997,
24464
+ cacheWrite: 5,
24187
24465
  },
24188
24466
  contextWindow: 1000000,
24189
24467
  maxTokens: 128000,
@@ -24530,6 +24808,78 @@ export const MODELS = {
24530
24808
  contextWindow: 1050000,
24531
24809
  maxTokens: 128000,
24532
24810
  },
24811
+ "openai/gpt-6.1-sol": {
24812
+ id: "openai/gpt-6.1-sol",
24813
+ name: "OpenAI: GPT-6.1 Sol",
24814
+ api: "openai-completions",
24815
+ provider: "openrouter",
24816
+ baseUrl: "https://openrouter.ai/api/v1",
24817
+ reasoning: true,
24818
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
24819
+ input: ["text", "image"],
24820
+ cost: {
24821
+ input: 2,
24822
+ output: 10,
24823
+ cacheRead: 0.09999999999999999,
24824
+ cacheWrite: 2.5,
24825
+ },
24826
+ contextWindow: 1050000,
24827
+ maxTokens: 128000,
24828
+ },
24829
+ "openai/gpt-6.1-sol-pro": {
24830
+ id: "openai/gpt-6.1-sol-pro",
24831
+ name: "OpenAI: GPT-6.1 Sol Pro",
24832
+ api: "openai-completions",
24833
+ provider: "openrouter",
24834
+ baseUrl: "https://openrouter.ai/api/v1",
24835
+ reasoning: true,
24836
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
24837
+ input: ["text", "image"],
24838
+ cost: {
24839
+ input: 2,
24840
+ output: 10,
24841
+ cacheRead: 0.09999999999999999,
24842
+ cacheWrite: 2.5,
24843
+ },
24844
+ contextWindow: 1050000,
24845
+ maxTokens: 128000,
24846
+ },
24847
+ "openai/gpt-6.1-sol-pro:batch": {
24848
+ id: "openai/gpt-6.1-sol-pro:batch",
24849
+ name: "OpenAI: GPT-6.1 Sol Pro (batch)",
24850
+ api: "openai-completions",
24851
+ provider: "openrouter",
24852
+ baseUrl: "https://openrouter.ai/api/v1",
24853
+ reasoning: true,
24854
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
24855
+ input: ["text", "image"],
24856
+ cost: {
24857
+ input: 1,
24858
+ output: 5,
24859
+ cacheRead: 0.049999999999999996,
24860
+ cacheWrite: 1.25,
24861
+ },
24862
+ contextWindow: 1050000,
24863
+ maxTokens: 128000,
24864
+ },
24865
+ "openai/gpt-6.1-sol:batch": {
24866
+ id: "openai/gpt-6.1-sol:batch",
24867
+ name: "OpenAI: GPT-6.1 Sol (batch)",
24868
+ api: "openai-completions",
24869
+ provider: "openrouter",
24870
+ baseUrl: "https://openrouter.ai/api/v1",
24871
+ reasoning: true,
24872
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
24873
+ input: ["text", "image"],
24874
+ cost: {
24875
+ input: 1,
24876
+ output: 5,
24877
+ cacheRead: 0.049999999999999996,
24878
+ cacheWrite: 1.25,
24879
+ },
24880
+ contextWindow: 1050000,
24881
+ maxTokens: 128000,
24882
+ },
24533
24883
  "openai/gpt-audio": {
24534
24884
  id: "openai/gpt-audio",
24535
24885
  name: "OpenAI: GPT Audio",
@@ -24591,13 +24941,31 @@ export const MODELS = {
24591
24941
  thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": null, "max": null, "ultra": null },
24592
24942
  input: ["text"],
24593
24943
  cost: {
24594
- input: 0.15,
24595
- output: 0.6,
24596
- cacheRead: 0.075,
24944
+ input: 0.037,
24945
+ output: 0.16999999999999998,
24946
+ cacheRead: 0,
24597
24947
  cacheWrite: 0,
24598
24948
  },
24599
24949
  contextWindow: 131072,
24600
- maxTokens: 65536,
24950
+ maxTokens: 117964,
24951
+ },
24952
+ "openai/gpt-oss-120b:batch": {
24953
+ id: "openai/gpt-oss-120b:batch",
24954
+ name: "OpenAI: gpt-oss-120b (batch)",
24955
+ api: "openai-completions",
24956
+ provider: "openrouter",
24957
+ baseUrl: "https://openrouter.ai/api/v1",
24958
+ reasoning: true,
24959
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": null, "max": null, "ultra": null },
24960
+ input: ["text"],
24961
+ cost: {
24962
+ input: 0.0296,
24963
+ output: 0.136,
24964
+ cacheRead: 0,
24965
+ cacheWrite: 0,
24966
+ },
24967
+ contextWindow: 131072,
24968
+ maxTokens: 117964,
24601
24969
  },
24602
24970
  "openai/gpt-oss-20b": {
24603
24971
  id: "openai/gpt-oss-20b",
@@ -24611,7 +24979,7 @@ export const MODELS = {
24611
24979
  cost: {
24612
24980
  input: 0.018,
24613
24981
  output: 0.09,
24614
- cacheRead: 0,
24982
+ cacheRead: 0.009,
24615
24983
  cacheWrite: 0,
24616
24984
  },
24617
24985
  contextWindow: 131072,
@@ -24876,6 +25244,24 @@ export const MODELS = {
24876
25244
  contextWindow: 200000,
24877
25245
  maxTokens: 4096,
24878
25246
  },
25247
+ "perceptron/perceptron-mk1.5": {
25248
+ id: "perceptron/perceptron-mk1.5",
25249
+ name: "Perceptron: Perceptron Mk1.5",
25250
+ api: "openai-completions",
25251
+ provider: "openrouter",
25252
+ baseUrl: "https://openrouter.ai/api/v1",
25253
+ reasoning: true,
25254
+ thinkingLevelMap: { "off": "none", "minimal": "minimal", "low": "low", "medium": "medium", "high": "high", "xhigh": null, "max": null, "ultra": null },
25255
+ input: ["text", "image"],
25256
+ cost: {
25257
+ input: 0.15,
25258
+ output: 1.5,
25259
+ cacheRead: 0,
25260
+ cacheWrite: 0,
25261
+ },
25262
+ contextWindow: 36864,
25263
+ maxTokens: 8192,
25264
+ },
24879
25265
  "poolside/laguna-s-2.1": {
24880
25266
  id: "poolside/laguna-s-2.1",
24881
25267
  name: "Poolside: Laguna S 2.1",
@@ -24956,7 +25342,7 @@ export const MODELS = {
24956
25342
  cost: {
24957
25343
  input: 0.075,
24958
25344
  output: 0.5,
24959
- cacheRead: 0,
25345
+ cacheRead: 0.0375,
24960
25346
  cacheWrite: 0,
24961
25347
  },
24962
25348
  contextWindow: 262144,
@@ -25111,13 +25497,13 @@ export const MODELS = {
25111
25497
  thinkingLevelMap: { "xhigh": "high", "max": "max" },
25112
25498
  input: ["text"],
25113
25499
  cost: {
25114
- input: 0.12,
25115
- output: 0.5,
25500
+ input: 0.13,
25501
+ output: 0.52,
25116
25502
  cacheRead: 0,
25117
25503
  cacheWrite: 0,
25118
25504
  },
25119
25505
  contextWindow: 131072,
25120
- maxTokens: 16384,
25506
+ maxTokens: 8192,
25121
25507
  },
25122
25508
  "qwen/qwen3-30b-a3b-instruct-2507": {
25123
25509
  id: "qwen/qwen3-30b-a3b-instruct-2507",
@@ -25319,13 +25705,13 @@ export const MODELS = {
25319
25705
  reasoning: false,
25320
25706
  input: ["text"],
25321
25707
  cost: {
25322
- input: 0.09,
25708
+ input: 0.09999999999999999,
25323
25709
  output: 1.1,
25324
- cacheRead: 0,
25710
+ cacheRead: 0.07,
25325
25711
  cacheWrite: 0,
25326
25712
  },
25327
25713
  contextWindow: 262144,
25328
- maxTokens: 16384,
25714
+ maxTokens: 235929,
25329
25715
  },
25330
25716
  "qwen/qwen3-next-80b-a3b-thinking": {
25331
25717
  id: "qwen/qwen3-next-80b-a3b-thinking",
@@ -25389,13 +25775,13 @@ export const MODELS = {
25389
25775
  reasoning: false,
25390
25776
  input: ["text", "image"],
25391
25777
  cost: {
25392
- input: 0.13,
25393
- output: 0.52,
25778
+ input: 0.15,
25779
+ output: 0.6,
25394
25780
  cacheRead: 0,
25395
25781
  cacheWrite: 0,
25396
25782
  },
25397
25783
  contextWindow: 262144,
25398
- maxTokens: 32768,
25784
+ maxTokens: 16384,
25399
25785
  },
25400
25786
  "qwen/qwen3-vl-30b-a3b-thinking": {
25401
25787
  id: "qwen/qwen3-vl-30b-a3b-thinking",
@@ -25513,13 +25899,13 @@ export const MODELS = {
25513
25899
  thinkingLevelMap: { "xhigh": "high", "max": "max" },
25514
25900
  input: ["text", "image"],
25515
25901
  cost: {
25516
- input: 0.3125,
25517
- output: 1.25,
25518
- cacheRead: 0.15625,
25902
+ input: 0.1625,
25903
+ output: 1.3,
25904
+ cacheRead: 0,
25519
25905
  cacheWrite: 0,
25520
25906
  },
25521
25907
  contextWindow: 262144,
25522
- maxTokens: 16384,
25908
+ maxTokens: 65536,
25523
25909
  },
25524
25910
  "qwen/qwen3.5-397b-a17b": {
25525
25911
  id: "qwen/qwen3.5-397b-a17b",
@@ -25622,12 +26008,12 @@ export const MODELS = {
25622
26008
  input: ["text", "image"],
25623
26009
  cost: {
25624
26010
  input: 0.32,
25625
- output: 2.7,
25626
- cacheRead: 0.15,
26011
+ output: 3.1999999999999997,
26012
+ cacheRead: 0,
25627
26013
  cacheWrite: 0,
25628
26014
  },
25629
26015
  contextWindow: 262144,
25630
- maxTokens: 262140,
26016
+ maxTokens: 81920,
25631
26017
  },
25632
26018
  "qwen/qwen3.6-35b-a3b": {
25633
26019
  id: "qwen/qwen3.6-35b-a3b",
@@ -25845,6 +26231,24 @@ export const MODELS = {
25845
26231
  contextWindow: 1000000,
25846
26232
  maxTokens: 131072,
25847
26233
  },
26234
+ "qwen/qwen3.8-max-prime": {
26235
+ id: "qwen/qwen3.8-max-prime",
26236
+ name: "Qwen: Qwen3.8 Max Prime",
26237
+ api: "openai-completions",
26238
+ provider: "openrouter",
26239
+ baseUrl: "https://openrouter.ai/api/v1",
26240
+ reasoning: true,
26241
+ thinkingLevelMap: { "xhigh": "xhigh", "max": null, "off": null, "minimal": "minimal", "low": "low", "medium": "medium", "high": "high", "ultra": null },
26242
+ input: ["text", "image"],
26243
+ cost: {
26244
+ input: 4,
26245
+ output: 12,
26246
+ cacheRead: 0.5,
26247
+ cacheWrite: 0,
26248
+ },
26249
+ contextWindow: 1000000,
26250
+ maxTokens: 131072,
26251
+ },
25848
26252
  "qwen/qwen3.8-omni-flash": {
25849
26253
  id: "qwen/qwen3.8-omni-flash",
25850
26254
  name: "Qwen: Qwen3.8 Omni Flash",
@@ -25986,6 +26390,24 @@ export const MODELS = {
25986
26390
  contextWindow: 131072,
25987
26391
  maxTokens: 16384,
25988
26392
  },
26393
+ "stealth/space-bunny-alpha": {
26394
+ id: "stealth/space-bunny-alpha",
26395
+ name: "Space Bunny Alpha",
26396
+ api: "openai-completions",
26397
+ provider: "openrouter",
26398
+ baseUrl: "https://openrouter.ai/api/v1",
26399
+ reasoning: true,
26400
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
26401
+ input: ["text", "image"],
26402
+ cost: {
26403
+ input: 0,
26404
+ output: 0,
26405
+ cacheRead: 0,
26406
+ cacheWrite: 0,
26407
+ },
26408
+ contextWindow: 1000000,
26409
+ maxTokens: 524288,
26410
+ },
25989
26411
  "stepfun/step-3.5-flash": {
25990
26412
  id: "stepfun/step-3.5-flash",
25991
26413
  name: "StepFun: Step 3.5 Flash",
@@ -26032,9 +26454,9 @@ export const MODELS = {
26032
26454
  thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": null, "ultra": null },
26033
26455
  input: ["text"],
26034
26456
  cost: {
26035
- input: 0.0825,
26036
- output: 0.33,
26037
- cacheRead: 0.020625,
26457
+ input: 0.13199999999999998,
26458
+ output: 0.5279999999999999,
26459
+ cacheRead: 0.032999999999999995,
26038
26460
  cacheWrite: 0,
26039
26461
  },
26040
26462
  contextWindow: 262144,
@@ -26091,7 +26513,7 @@ export const MODELS = {
26091
26513
  cacheRead: 0.16999999999999998,
26092
26514
  cacheWrite: 0,
26093
26515
  },
26094
- contextWindow: 1048576,
26516
+ contextWindow: 524288,
26095
26517
  maxTokens: 471859,
26096
26518
  },
26097
26519
  "thinkingmachines/inkling-small": {
@@ -26109,7 +26531,7 @@ export const MODELS = {
26109
26531
  cacheRead: 0.09999999999999999,
26110
26532
  cacheWrite: 0,
26111
26533
  },
26112
- contextWindow: 1048576,
26534
+ contextWindow: 524288,
26113
26535
  maxTokens: 262144,
26114
26536
  },
26115
26537
  "thinkingmachines/inkling-small:free": {
@@ -26165,6 +26587,24 @@ export const MODELS = {
26165
26587
  contextWindow: 262144,
26166
26588
  maxTokens: 131072,
26167
26589
  },
26590
+ "upstage/solar-mini4": {
26591
+ id: "upstage/solar-mini4",
26592
+ name: "Upstage: Solar Mini 4",
26593
+ api: "openai-completions",
26594
+ provider: "openrouter",
26595
+ baseUrl: "https://openrouter.ai/api/v1",
26596
+ reasoning: true,
26597
+ thinkingLevelMap: { "off": "none", "minimal": "minimal", "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
26598
+ input: ["text"],
26599
+ cost: {
26600
+ input: 0.049999999999999996,
26601
+ output: 0.19999999999999998,
26602
+ cacheRead: 0.005,
26603
+ cacheWrite: 0,
26604
+ },
26605
+ contextWindow: 524288,
26606
+ maxTokens: 131072,
26607
+ },
26168
26608
  "upstage/solar-pro-3": {
26169
26609
  id: "upstage/solar-pro-3",
26170
26610
  name: "Upstage: Solar Pro 3",
@@ -26300,9 +26740,9 @@ export const MODELS = {
26300
26740
  thinkingLevelMap: { "xhigh": "xhigh", "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "max": null, "ultra": null },
26301
26741
  input: ["text", "image"],
26302
26742
  cost: {
26303
- input: 1.5999999999999999,
26304
- output: 4.8,
26305
- cacheRead: 0.39999999999999997,
26743
+ input: 2,
26744
+ output: 6,
26745
+ cacheRead: 0.5,
26306
26746
  cacheWrite: 0,
26307
26747
  },
26308
26748
  contextWindow: 500000,
@@ -26391,7 +26831,7 @@ export const MODELS = {
26391
26831
  cacheRead: 0.0036,
26392
26832
  cacheWrite: 0,
26393
26833
  },
26394
- contextWindow: 1048576,
26834
+ contextWindow: 1050000,
26395
26835
  maxTokens: 131072,
26396
26836
  },
26397
26837
  "xiaomi/mimo-v2.6-pro-ultraspeed": {
@@ -26505,9 +26945,9 @@ export const MODELS = {
26505
26945
  reasoning: true,
26506
26946
  input: ["text"],
26507
26947
  cost: {
26508
- input: 0.39999999999999997,
26509
- output: 1.75,
26510
- cacheRead: 0.08,
26948
+ input: 0.6,
26949
+ output: 2.2,
26950
+ cacheRead: 0.11,
26511
26951
  cacheWrite: 0,
26512
26952
  },
26513
26953
  contextWindow: 204800,
@@ -26573,13 +27013,13 @@ export const MODELS = {
26573
27013
  reasoning: true,
26574
27014
  input: ["text"],
26575
27015
  cost: {
26576
- input: 0.966,
26577
- output: 3.036,
26578
- cacheRead: 0.1794,
27016
+ input: 1.4,
27017
+ output: 4.4,
27018
+ cacheRead: 0.26,
26579
27019
  cacheWrite: 0,
26580
27020
  },
26581
27021
  contextWindow: 204800,
26582
- maxTokens: 128000,
27022
+ maxTokens: 131072,
26583
27023
  },
26584
27024
  "z-ai/glm-5.2": {
26585
27025
  id: "z-ai/glm-5.2",
@@ -26591,13 +27031,13 @@ export const MODELS = {
26591
27031
  thinkingLevelMap: { "off": "none", "minimal": null, "low": null, "medium": null, "high": "high", "xhigh": "xhigh", "max": null, "ultra": null },
26592
27032
  input: ["text"],
26593
27033
  cost: {
26594
- input: 0.6496,
26595
- output: 2.0416,
26596
- cacheRead: 0.12064,
27034
+ input: 0.2087,
27035
+ output: 3.9899999999999998,
27036
+ cacheRead: 0.167,
26597
27037
  cacheWrite: 0,
26598
27038
  },
26599
27039
  contextWindow: 1048576,
26600
- maxTokens: 131072,
27040
+ maxTokens: 943718,
26601
27041
  },
26602
27042
  "z-ai/glm-5.3": {
26603
27043
  id: "z-ai/glm-5.3",
@@ -26609,13 +27049,13 @@ export const MODELS = {
26609
27049
  thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
26610
27050
  input: ["text"],
26611
27051
  cost: {
26612
- input: 0.6537999999999999,
26613
- output: 2.0547999999999997,
26614
- cacheRead: 0.12142000000000001,
27052
+ input: 1.4,
27053
+ output: 4.4,
27054
+ cacheRead: 0.26,
26615
27055
  cacheWrite: 0,
26616
27056
  },
26617
- contextWindow: 1310720,
26618
- maxTokens: 131072,
27057
+ contextWindow: 1048576,
27058
+ maxTokens: 943717,
26619
27059
  },
26620
27060
  "z-ai/glm-5.3-flash": {
26621
27061
  id: "z-ai/glm-5.3-flash",
@@ -26629,11 +27069,11 @@ export const MODELS = {
26629
27069
  cost: {
26630
27070
  input: 0.15,
26631
27071
  output: 0.5,
26632
- cacheRead: 0.049999999999999996,
27072
+ cacheRead: 0.03,
26633
27073
  cacheWrite: 0,
26634
27074
  },
26635
- contextWindow: 1310720,
26636
- maxTokens: 943718,
27075
+ contextWindow: 1048576,
27076
+ maxTokens: 943717,
26637
27077
  },
26638
27078
  "z-ai/glm-5.3-flash:batch": {
26639
27079
  id: "z-ai/glm-5.3-flash:batch",
@@ -26665,12 +27105,30 @@ export const MODELS = {
26665
27105
  cost: {
26666
27106
  input: 0.37,
26667
27107
  output: 1.25,
26668
- cacheRead: 0.075,
27108
+ cacheRead: 0.09,
26669
27109
  cacheWrite: 0,
26670
27110
  },
26671
27111
  contextWindow: 1048576,
26672
27112
  maxTokens: 131072,
26673
27113
  },
27114
+ "z-ai/glm-5.3-prime": {
27115
+ id: "z-ai/glm-5.3-prime",
27116
+ name: "Z.ai: GLM 5.3 Prime",
27117
+ api: "openai-completions",
27118
+ provider: "openrouter",
27119
+ baseUrl: "https://openrouter.ai/api/v1",
27120
+ reasoning: true,
27121
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
27122
+ input: ["text"],
27123
+ cost: {
27124
+ input: 2.8,
27125
+ output: 8.8,
27126
+ cacheRead: 0.56,
27127
+ cacheWrite: 0,
27128
+ },
27129
+ contextWindow: 1000000,
27130
+ maxTokens: 131072,
27131
+ },
26674
27132
  "z-ai/glm-5.3:batch": {
26675
27133
  id: "z-ai/glm-5.3:batch",
26676
27134
  name: "Z.ai: GLM 5.3 (batch)",
@@ -26681,9 +27139,9 @@ export const MODELS = {
26681
27139
  thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
26682
27140
  input: ["text"],
26683
27141
  cost: {
26684
- input: 0.72,
26685
- output: 2.4,
26686
- cacheRead: 0.12,
27142
+ input: 0.44999999999999996,
27143
+ output: 2,
27144
+ cacheRead: 0.09999999999999999,
26687
27145
  cacheWrite: 0,
26688
27146
  },
26689
27147
  contextWindow: 1048576,
@@ -26766,7 +27224,7 @@ export const MODELS = {
26766
27224
  provider: "openrouter",
26767
27225
  baseUrl: "https://openrouter.ai/api/v1",
26768
27226
  reasoning: true,
26769
- thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
27227
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
26770
27228
  input: ["text", "image"],
26771
27229
  cost: {
26772
27230
  input: 2,
@@ -26787,9 +27245,9 @@ export const MODELS = {
26787
27245
  thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
26788
27246
  input: ["text", "image"],
26789
27247
  cost: {
26790
- input: 0.12,
26791
- output: 0.48,
26792
- cacheRead: 0.0036,
27248
+ input: 0.02,
27249
+ output: 0.6,
27250
+ cacheRead: 0.02,
26793
27251
  cacheWrite: 0,
26794
27252
  },
26795
27253
  contextWindow: 1048576,
@@ -26805,9 +27263,9 @@ export const MODELS = {
26805
27263
  thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
26806
27264
  input: ["text"],
26807
27265
  cost: {
26808
- input: 0.39996,
26809
- output: 1.19988,
26810
- cacheRead: 0.012726,
27266
+ input: 0.15,
27267
+ output: 3.5,
27268
+ cacheRead: 0.15,
26811
27269
  cacheWrite: 0,
26812
27270
  },
26813
27271
  contextWindow: 1048576,
@@ -26824,12 +27282,12 @@ export const MODELS = {
26824
27282
  thinkingLevelMap: { "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "off": "none", "max": "max", "ultra": null },
26825
27283
  input: ["text"],
26826
27284
  cost: {
26827
- input: 0.03,
26828
- output: 0.7999999999999999,
26829
- cacheRead: 0.008,
27285
+ input: 0.012,
27286
+ output: 1.25,
27287
+ cacheRead: 0.01,
26830
27288
  cacheWrite: 0,
26831
27289
  },
26832
- contextWindow: 1310720,
27290
+ contextWindow: 1048576,
26833
27291
  maxTokens: 943718,
26834
27292
  },
26835
27293
  "~google/gemini-flash-latest": {
@@ -26878,9 +27336,9 @@ export const MODELS = {
26878
27336
  thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
26879
27337
  input: ["text", "image"],
26880
27338
  cost: {
26881
- input: 1.4989,
26882
- output: 10.758,
26883
- cacheRead: 0.3,
27339
+ input: 0.39999999999999997,
27340
+ output: 9,
27341
+ cacheRead: 0.39999999999999997,
26884
27342
  cacheWrite: 0,
26885
27343
  },
26886
27344
  contextWindow: 1048576,
@@ -26947,12 +27405,12 @@ export const MODELS = {
26947
27405
  provider: "openrouter",
26948
27406
  baseUrl: "https://openrouter.ai/api/v1",
26949
27407
  reasoning: true,
26950
- thinkingLevelMap: { "off": "none", "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
27408
+ thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max", "ultra": null },
26951
27409
  input: ["text", "image"],
26952
27410
  cost: {
26953
27411
  input: 2,
26954
27412
  output: 10,
26955
- cacheRead: 0.19999999999999998,
27413
+ cacheRead: 0.09999999999999999,
26956
27414
  cacheWrite: 2.5,
26957
27415
  },
26958
27416
  contextWindow: 1050000,
@@ -26986,9 +27444,9 @@ export const MODELS = {
26986
27444
  thinkingLevelMap: { "xhigh": "xhigh", "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "max": null, "ultra": null },
26987
27445
  input: ["text", "image"],
26988
27446
  cost: {
26989
- input: 1.5999999999999999,
26990
- output: 4.8,
26991
- cacheRead: 0.39999999999999997,
27447
+ input: 2,
27448
+ output: 6,
27449
+ cacheRead: 0.5,
26992
27450
  cacheWrite: 0,
26993
27451
  },
26994
27452
  contextWindow: 500000,
@@ -27004,12 +27462,12 @@ export const MODELS = {
27004
27462
  thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
27005
27463
  input: ["text", "image"],
27006
27464
  cost: {
27007
- input: 0.075,
27008
- output: 0.25,
27009
- cacheRead: 0.015,
27465
+ input: 0.02,
27466
+ output: 0.3,
27467
+ cacheRead: 0.01,
27010
27468
  cacheWrite: 0,
27011
27469
  },
27012
- contextWindow: 1310720,
27470
+ contextWindow: 1048576,
27013
27471
  maxTokens: 943718,
27014
27472
  },
27015
27473
  "~z-ai/glm-latest": {
@@ -27022,12 +27480,12 @@ export const MODELS = {
27022
27480
  thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": null, "high": "high", "xhigh": null, "max": "max", "ultra": null },
27023
27481
  input: ["text"],
27024
27482
  cost: {
27025
- input: 0.6537999999999999,
27026
- output: 2.0547999999999997,
27027
- cacheRead: 0.12142000000000001,
27483
+ input: 0.12,
27484
+ output: 1.5,
27485
+ cacheRead: 0.12,
27028
27486
  cacheWrite: 0,
27029
27487
  },
27030
- contextWindow: 1310720,
27488
+ contextWindow: 1048576,
27031
27489
  maxTokens: 131072,
27032
27490
  },
27033
27491
  },
@@ -27048,7 +27506,7 @@ export const MODELS = {
27048
27506
  cacheRead: 0.06,
27049
27507
  cacheWrite: 0,
27050
27508
  },
27051
- contextWindow: 202752,
27509
+ contextWindow: 196608,
27052
27510
  maxTokens: 131072,
27053
27511
  },
27054
27512
  "MiniMaxAI/MiniMax-M3": {
@@ -27160,7 +27618,7 @@ export const MODELS = {
27160
27618
  cacheRead: 0.03,
27161
27619
  cacheWrite: 0,
27162
27620
  },
27163
- contextWindow: 1000000,
27621
+ contextWindow: 1048576,
27164
27622
  maxTokens: 384000,
27165
27623
  },
27166
27624
  "deepseek-ai/DeepSeek-V4-Pro": {
@@ -27257,44 +27715,6 @@ export const MODELS = {
27257
27715
  contextWindow: 131072,
27258
27716
  maxTokens: 131072,
27259
27717
  },
27260
- "moonshotai/Kimi-K2.6": {
27261
- id: "moonshotai/Kimi-K2.6",
27262
- name: "Kimi K2.6",
27263
- api: "openai-completions",
27264
- provider: "together",
27265
- baseUrl: "https://api.together.ai/v1",
27266
- compat: { "supportsStore": false, "supportsDeveloperRole": false, "supportsReasoningEffort": false, "maxTokensField": "max_tokens", "supportsStrictMode": false, "supportsLongCacheRetention": false, "thinkingFormat": "together" },
27267
- reasoning: true,
27268
- thinkingLevelMap: { "minimal": null, "low": null, "medium": null },
27269
- input: ["text", "image"],
27270
- cost: {
27271
- input: 1.2,
27272
- output: 4.5,
27273
- cacheRead: 0.2,
27274
- cacheWrite: 0,
27275
- },
27276
- contextWindow: 262144,
27277
- maxTokens: 131000,
27278
- },
27279
- "moonshotai/Kimi-K2.7-Code": {
27280
- id: "moonshotai/Kimi-K2.7-Code",
27281
- name: "Kimi K2.7 Code",
27282
- api: "openai-completions",
27283
- provider: "together",
27284
- baseUrl: "https://api.together.ai/v1",
27285
- compat: { "supportsStore": false, "supportsDeveloperRole": false, "supportsReasoningEffort": false, "maxTokensField": "max_tokens", "supportsStrictMode": false, "supportsLongCacheRetention": false, "thinkingFormat": "together" },
27286
- reasoning: true,
27287
- thinkingLevelMap: { "minimal": null, "low": null, "medium": null },
27288
- input: ["text"],
27289
- cost: {
27290
- input: 0.95,
27291
- output: 4,
27292
- cacheRead: 0.19,
27293
- cacheWrite: 0,
27294
- },
27295
- contextWindow: 262144,
27296
- maxTokens: 131072,
27297
- },
27298
27718
  "moonshotai/Kimi-K3": {
27299
27719
  id: "moonshotai/Kimi-K3",
27300
27720
  name: "Kimi K3",
@@ -27406,7 +27826,7 @@ export const MODELS = {
27406
27826
  cacheRead: 0.26,
27407
27827
  cacheWrite: 0,
27408
27828
  },
27409
- contextWindow: 512000,
27829
+ contextWindow: 1048575,
27410
27830
  maxTokens: 164000,
27411
27831
  },
27412
27832
  "zai-org/GLM-5.3": {
@@ -27793,7 +28213,7 @@ export const MODELS = {
27793
28213
  input: ["text", "image"],
27794
28214
  cost: {
27795
28215
  input: 0.39999999999999997,
27796
- output: 2.5,
28216
+ output: 2.4,
27797
28217
  cacheRead: 0.04,
27798
28218
  cacheWrite: 0.5,
27799
28219
  },
@@ -27980,6 +28400,24 @@ export const MODELS = {
27980
28400
  contextWindow: 991000,
27981
28401
  maxTokens: 128000,
27982
28402
  },
28403
+ "alibaba/qwen3.8-max-prime": {
28404
+ id: "alibaba/qwen3.8-max-prime",
28405
+ name: "Qwen 3.8 Max Prime",
28406
+ api: "anthropic-messages",
28407
+ provider: "vercel-ai-gateway",
28408
+ baseUrl: "https://ai-gateway.vercel.sh",
28409
+ reasoning: true,
28410
+ thinkingLevelMap: { "xhigh": "high", "max": "max" },
28411
+ input: ["text", "image"],
28412
+ cost: {
28413
+ input: 4,
28414
+ output: 12,
28415
+ cacheRead: 0.5,
28416
+ cacheWrite: 5,
28417
+ },
28418
+ contextWindow: 1000000,
28419
+ maxTokens: 131072,
28420
+ },
27983
28421
  "alibaba/qwen3.8-omni-flash": {
27984
28422
  id: "alibaba/qwen3.8-omni-flash",
27985
28423
  name: "Qwen 3.8 Omni Flash",
@@ -28153,7 +28591,7 @@ export const MODELS = {
28153
28591
  cacheWrite: 18.75,
28154
28592
  },
28155
28593
  contextWindow: 200000,
28156
- maxTokens: 8192,
28594
+ maxTokens: 32000,
28157
28595
  },
28158
28596
  "anthropic/claude-opus-4.5": {
28159
28597
  id: "anthropic/claude-opus-4.5",
@@ -28339,7 +28777,7 @@ export const MODELS = {
28339
28777
  cacheWrite: 3.75,
28340
28778
  },
28341
28779
  contextWindow: 1000000,
28342
- maxTokens: 8192,
28780
+ maxTokens: 64000,
28343
28781
  },
28344
28782
  "anthropic/claude-sonnet-4.5": {
28345
28783
  id: "anthropic/claude-sonnet-4.5",
@@ -28608,9 +29046,9 @@ export const MODELS = {
28608
29046
  reasoning: true,
28609
29047
  input: ["text", "image"],
28610
29048
  cost: {
28611
- input: 0.22,
28612
- output: 0.66,
28613
- cacheRead: 0.007,
29049
+ input: 0.21559999999999999,
29050
+ output: 0.6468,
29051
+ cacheRead: 0.0068,
28614
29052
  cacheWrite: 0,
28615
29053
  },
28616
29054
  contextWindow: 1048576,
@@ -28669,6 +29107,23 @@ export const MODELS = {
28669
29107
  contextWindow: 1048576,
28670
29108
  maxTokens: 32768,
28671
29109
  },
29110
+ "fireworks/ember-1": {
29111
+ id: "fireworks/ember-1",
29112
+ name: "Ember-1",
29113
+ api: "anthropic-messages",
29114
+ provider: "vercel-ai-gateway",
29115
+ baseUrl: "https://ai-gateway.vercel.sh",
29116
+ reasoning: true,
29117
+ input: ["text", "image"],
29118
+ cost: {
29119
+ input: 3,
29120
+ output: 15,
29121
+ cacheRead: 0.3,
29122
+ cacheWrite: 0,
29123
+ },
29124
+ contextWindow: 1048576,
29125
+ maxTokens: 1048576,
29126
+ },
28672
29127
  "google/gemini-2.5-flash": {
28673
29128
  id: "google/gemini-2.5-flash",
28674
29129
  name: "Gemini 2.5 Flash",
@@ -28684,7 +29139,7 @@ export const MODELS = {
28684
29139
  cacheWrite: 0,
28685
29140
  },
28686
29141
  contextWindow: 1000000,
28687
- maxTokens: 65536,
29142
+ maxTokens: 65535,
28688
29143
  },
28689
29144
  "google/gemini-2.5-flash-lite": {
28690
29145
  id: "google/gemini-2.5-flash-lite",
@@ -28701,7 +29156,7 @@ export const MODELS = {
28701
29156
  cacheWrite: 0,
28702
29157
  },
28703
29158
  contextWindow: 1048576,
28704
- maxTokens: 65536,
29159
+ maxTokens: 65535,
28705
29160
  },
28706
29161
  "google/gemini-2.5-pro": {
28707
29162
  id: "google/gemini-2.5-pro",
@@ -28718,7 +29173,7 @@ export const MODELS = {
28718
29173
  cacheWrite: 0,
28719
29174
  },
28720
29175
  contextWindow: 1048576,
28721
- maxTokens: 65536,
29176
+ maxTokens: 65535,
28722
29177
  },
28723
29178
  "google/gemini-3-flash": {
28724
29179
  id: "google/gemini-3-flash",
@@ -28837,7 +29292,7 @@ export const MODELS = {
28837
29292
  cacheWrite: 0,
28838
29293
  },
28839
29294
  contextWindow: 1000000,
28840
- maxTokens: 65536,
29295
+ maxTokens: 65535,
28841
29296
  },
28842
29297
  "google/gemini-3.8-flash": {
28843
29298
  id: "google/gemini-3.8-flash",
@@ -28854,7 +29309,7 @@ export const MODELS = {
28854
29309
  cacheWrite: 0,
28855
29310
  },
28856
29311
  contextWindow: 1000000,
28857
- maxTokens: 65536,
29312
+ maxTokens: 65535,
28858
29313
  },
28859
29314
  "google/gemma-4-26b-a4b-it": {
28860
29315
  id: "google/gemma-4-26b-a4b-it",
@@ -28967,26 +29422,9 @@ export const MODELS = {
28967
29422
  reasoning: true,
28968
29423
  input: ["text"],
28969
29424
  cost: {
28970
- input: 0,
28971
- output: 0,
28972
- cacheRead: 0,
28973
- cacheWrite: 0,
28974
- },
28975
- contextWindow: 256000,
28976
- maxTokens: 32000,
28977
- },
28978
- "inclusionai/ling-3.0-flash-fin-free": {
28979
- id: "inclusionai/ling-3.0-flash-fin-free",
28980
- name: "Ling 3.0 Flash Fin (Free)",
28981
- api: "anthropic-messages",
28982
- provider: "vercel-ai-gateway",
28983
- baseUrl: "https://ai-gateway.vercel.sh",
28984
- reasoning: true,
28985
- input: ["text"],
28986
- cost: {
28987
- input: 0,
28988
- output: 0,
28989
- cacheRead: 0,
29425
+ input: 0.075,
29426
+ output: 0.22,
29427
+ cacheRead: 0.015,
28990
29428
  cacheWrite: 0,
28991
29429
  },
28992
29430
  contextWindow: 256000,
@@ -29034,31 +29472,48 @@ export const MODELS = {
29034
29472
  baseUrl: "https://ai-gateway.vercel.sh",
29035
29473
  reasoning: true,
29036
29474
  input: ["text", "image"],
29475
+ cost: {
29476
+ input: 0.075,
29477
+ output: 0.22,
29478
+ cacheRead: 0.015,
29479
+ cacheWrite: 0,
29480
+ },
29481
+ contextWindow: 256000,
29482
+ maxTokens: 32000,
29483
+ },
29484
+ "inclusionai/ling-3.1-flash": {
29485
+ id: "inclusionai/ling-3.1-flash",
29486
+ name: "Ling 3.1 Flash",
29487
+ api: "anthropic-messages",
29488
+ provider: "vercel-ai-gateway",
29489
+ baseUrl: "https://ai-gateway.vercel.sh",
29490
+ reasoning: true,
29491
+ input: ["text"],
29037
29492
  cost: {
29038
29493
  input: 0,
29039
29494
  output: 0,
29040
29495
  cacheRead: 0,
29041
29496
  cacheWrite: 0,
29042
29497
  },
29043
- contextWindow: 256000,
29044
- maxTokens: 32000,
29498
+ contextWindow: 262144,
29499
+ maxTokens: 32768,
29045
29500
  },
29046
- "inclusionai/ling-3.0-flash-vl-free": {
29047
- id: "inclusionai/ling-3.0-flash-vl-free",
29048
- name: "Ling 3.0 Flash VL (Free)",
29501
+ "inclusionai/ling-3.1-flash-free": {
29502
+ id: "inclusionai/ling-3.1-flash-free",
29503
+ name: "Ling 3.1 Flash (Free)",
29049
29504
  api: "anthropic-messages",
29050
29505
  provider: "vercel-ai-gateway",
29051
29506
  baseUrl: "https://ai-gateway.vercel.sh",
29052
29507
  reasoning: true,
29053
- input: ["text", "image"],
29508
+ input: ["text"],
29054
29509
  cost: {
29055
29510
  input: 0,
29056
29511
  output: 0,
29057
29512
  cacheRead: 0,
29058
29513
  cacheWrite: 0,
29059
29514
  },
29060
- contextWindow: 256000,
29061
- maxTokens: 32000,
29515
+ contextWindow: 262144,
29516
+ maxTokens: 32768,
29062
29517
  },
29063
29518
  "interfaze/interfaze-beta": {
29064
29519
  id: "interfaze/interfaze-beta",
@@ -29077,6 +29532,23 @@ export const MODELS = {
29077
29532
  contextWindow: 1000000,
29078
29533
  maxTokens: 32000,
29079
29534
  },
29535
+ "meituan/longcat-2.5-preview": {
29536
+ id: "meituan/longcat-2.5-preview",
29537
+ name: "LongCat 2.5 Preview",
29538
+ api: "anthropic-messages",
29539
+ provider: "vercel-ai-gateway",
29540
+ baseUrl: "https://ai-gateway.vercel.sh",
29541
+ reasoning: true,
29542
+ input: ["text", "image"],
29543
+ cost: {
29544
+ input: 0.3,
29545
+ output: 1.2,
29546
+ cacheRead: 0.006,
29547
+ cacheWrite: 0,
29548
+ },
29549
+ contextWindow: 1048576,
29550
+ maxTokens: 131072,
29551
+ },
29080
29552
  "meta/llama-3.1-70b": {
29081
29553
  id: "meta/llama-3.1-70b",
29082
29554
  name: "Llama 3.1 70B Instruct",
@@ -29632,7 +30104,7 @@ export const MODELS = {
29632
30104
  cost: {
29633
30105
  input: 0.95,
29634
30106
  output: 4,
29635
- cacheRead: 0.16,
30107
+ cacheRead: 0.19,
29636
30108
  cacheWrite: 0,
29637
30109
  },
29638
30110
  contextWindow: 256000,
@@ -30679,6 +31151,40 @@ export const MODELS = {
30679
31151
  contextWindow: 1050000,
30680
31152
  maxTokens: 128000,
30681
31153
  },
31154
+ "openai/gpt-6.1-sol": {
31155
+ id: "openai/gpt-6.1-sol",
31156
+ name: "GPT-6.1 Sol",
31157
+ api: "anthropic-messages",
31158
+ provider: "vercel-ai-gateway",
31159
+ baseUrl: "https://ai-gateway.vercel.sh",
31160
+ reasoning: true,
31161
+ input: ["text", "image"],
31162
+ cost: {
31163
+ input: 2,
31164
+ output: 10,
31165
+ cacheRead: 0.09999999999999999,
31166
+ cacheWrite: 2.5,
31167
+ },
31168
+ contextWindow: 1050000,
31169
+ maxTokens: 128000,
31170
+ },
31171
+ "openai/gpt-6.1-sol-fast": {
31172
+ id: "openai/gpt-6.1-sol-fast",
31173
+ name: "GPT-6.1 Sol (Fast)",
31174
+ api: "anthropic-messages",
31175
+ provider: "vercel-ai-gateway",
31176
+ baseUrl: "https://ai-gateway.vercel.sh",
31177
+ reasoning: true,
31178
+ input: ["text", "image"],
31179
+ cost: {
31180
+ input: 4,
31181
+ output: 20,
31182
+ cacheRead: 0.19999999999999998,
31183
+ cacheWrite: 5,
31184
+ },
31185
+ contextWindow: 1050000,
31186
+ maxTokens: 128000,
31187
+ },
30682
31188
  "openai/gpt-oss-120b": {
30683
31189
  id: "openai/gpt-oss-120b",
30684
31190
  name: "GPT OSS 120B",
@@ -30690,7 +31196,7 @@ export const MODELS = {
30690
31196
  cost: {
30691
31197
  input: 0.09999999999999999,
30692
31198
  output: 0.5,
30693
- cacheRead: 0,
31199
+ cacheRead: 0.09999999999999999,
30694
31200
  cacheWrite: 0,
30695
31201
  },
30696
31202
  contextWindow: 131072,
@@ -31201,9 +31707,9 @@ export const MODELS = {
31201
31707
  thinkingLevelMap: { "xhigh": "xhigh" },
31202
31708
  input: ["text", "image"],
31203
31709
  cost: {
31204
- input: 1.2,
31205
- output: 3.5999999999999996,
31206
- cacheRead: 0.3,
31710
+ input: 2,
31711
+ output: 6,
31712
+ cacheRead: 0.5,
31207
31713
  cacheWrite: 0,
31208
31714
  },
31209
31715
  contextWindow: 500000,
@@ -31320,7 +31826,7 @@ export const MODELS = {
31320
31826
  reasoning: true,
31321
31827
  input: ["text", "image"],
31322
31828
  cost: {
31323
- input: 0.5,
31829
+ input: 0.44999999999999996,
31324
31830
  output: 1.2,
31325
31831
  cacheRead: 0.09999999999999999,
31326
31832
  cacheWrite: 0,
@@ -31492,7 +31998,7 @@ export const MODELS = {
31492
31998
  cost: {
31493
31999
  input: 0.6,
31494
32000
  output: 2.2,
31495
- cacheRead: 0.12,
32001
+ cacheRead: 0,
31496
32002
  cacheWrite: 0,
31497
32003
  },
31498
32004
  contextWindow: 200000,
@@ -31611,9 +32117,9 @@ export const MODELS = {
31611
32117
  thinkingLevelMap: { "off": null, "minimal": null, "low": "low", "medium": "medium", "high": "high", "xhigh": "xhigh", "max": "max" },
31612
32118
  input: ["text"],
31613
32119
  cost: {
31614
- input: 2.0999999999999996,
31615
- output: 6.6000000000000005,
31616
- cacheRead: 0.21,
32120
+ input: 2.8,
32121
+ output: 8.8,
32122
+ cacheRead: 0.56,
31617
32123
  cacheWrite: 0,
31618
32124
  },
31619
32125
  contextWindow: 1000000,