@llmgateway/models 1.210.0 → 1.211.1

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 (8) hide show
  1. package/dist/models/alibaba.d.ts +159 -6
  2. package/dist/models/alibaba.js +208 -11
  3. package/dist/models/alibaba.js.map +1 -1
  4. package/dist/models/google.d.ts +6 -0
  5. package/dist/models/google.js +6 -0
  6. package/dist/models/google.js.map +1 -1
  7. package/dist/models.d.ts +165 -6
  8. package/package.json +1 -1
package/dist/models/alibaba.d.ts CHANGED
@@ -367,6 +367,7 @@ export declare const alibabaModels: [{
367
367
  readonly vision: false;
368
368
  readonly tools: true;
369
369
  readonly jsonOutput: true;
370
+ readonly deactivatedAt: Date;
370
371
  }];
371
372
  }, {
372
373
  readonly id: "qwen-qwq-32b";
@@ -1052,6 +1053,7 @@ export declare const alibabaModels: [{
1052
1053
  readonly vision: true;
1053
1054
  readonly tools: true;
1054
1055
  readonly jsonOutput: true;
1056
+ readonly deactivatedAt: Date;
1055
1057
  }, {
1056
1058
  readonly providerId: "novita";
1057
1059
  readonly externalId: "qwen/qwen3-max";
@@ -1191,6 +1193,62 @@ export declare const alibabaModels: [{
1191
1193
  readonly jsonOutput: true;
1192
1194
  readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "stream", "tools", "response_format", "reasoning_effort"];
1193
1195
  }];
1196
+ }, {
1197
+ readonly id: "qwen3.7-flash";
1198
+ readonly name: "Qwen3.7 Flash";
1199
+ readonly description: "Fast, cost-effective Qwen3.7 model with hybrid reasoning and tool calling for high-volume tasks.";
1200
+ readonly family: "alibaba";
1201
+ readonly releasedAt: Date;
1202
+ readonly providers: [{
1203
+ readonly providerId: "alibaba";
1204
+ readonly externalId: "qwen3.7-flash";
1205
+ readonly inputPrice: "0.03e-6";
1206
+ readonly outputPrice: "0.13e-6";
1207
+ readonly cachedInputPrice: "0.006e-6";
1208
+ readonly cacheReadInputPrice: "0.003e-6";
1209
+ readonly cacheWriteInputPrice: "0.0375e-6";
1210
+ readonly pricingTiers: [{
1211
+ readonly name: "Up to 32K";
1212
+ readonly upToTokens: 32000;
1213
+ readonly inputPrice: "0.03e-6";
1214
+ readonly outputPrice: "0.13e-6";
1215
+ readonly cachedInputPrice: "0.006e-6";
1216
+ readonly cacheReadInputPrice: "0.003e-6";
1217
+ readonly cacheWriteInputPrice: "0.0375e-6";
1218
+ }, {
1219
+ readonly name: "32K-256K";
1220
+ readonly upToTokens: 256000;
1221
+ readonly inputPrice: "0.10e-6";
1222
+ readonly outputPrice: "0.40e-6";
1223
+ readonly cachedInputPrice: "0.02e-6";
1224
+ readonly cacheReadInputPrice: "0.01e-6";
1225
+ readonly cacheWriteInputPrice: "0.125e-6";
1226
+ }, {
1227
+ readonly name: "256K-1M";
1228
+ readonly upToTokens: 1000000;
1229
+ readonly inputPrice: "0.20e-6";
1230
+ readonly outputPrice: "0.80e-6";
1231
+ readonly cachedInputPrice: "0.04e-6";
1232
+ readonly cacheReadInputPrice: "0.02e-6";
1233
+ readonly cacheWriteInputPrice: "0.25e-6";
1234
+ }];
1235
+ readonly regions: [{
1236
+ readonly id: "singapore";
1237
+ }, {
1238
+ readonly id: "cn-beijing";
1239
+ }];
1240
+ readonly requestPrice: "0";
1241
+ readonly contextSize: 1000000;
1242
+ readonly maxOutput: 65536;
1243
+ readonly reasoning: true;
1244
+ readonly reasoningMaxTokens: true;
1245
+ readonly reasoningOutput: "omit";
1246
+ readonly streaming: true;
1247
+ readonly vision: true;
1248
+ readonly tools: true;
1249
+ readonly jsonOutput: true;
1250
+ readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "stream", "tools", "response_format", "reasoning_effort"];
1251
+ }];
1194
1252
  }, {
1195
1253
  readonly id: "qwen3-coder-next";
1196
1254
  readonly name: "Qwen3 Coder Next";
@@ -1232,6 +1290,21 @@ export declare const alibabaModels: [{
1232
1290
  readonly tools: true;
1233
1291
  readonly jsonOutput: false;
1234
1292
  readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "response_format", "tools", "tool_choice"];
1293
+ }, {
1294
+ readonly providerId: "deepinfra";
1295
+ readonly externalId: "Qwen/Qwen3-VL-30B-A3B-Instruct";
1296
+ readonly inputPrice: "0.15e-6";
1297
+ readonly outputPrice: "0.60e-6";
1298
+ readonly requestPrice: "0";
1299
+ readonly contextSize: 262144;
1300
+ readonly maxOutput: 32768;
1301
+ readonly quantization: "fp8";
1302
+ readonly streaming: true;
1303
+ readonly vision: true;
1304
+ readonly tools: true;
1305
+ readonly supportedToolChoices: ["auto", "none", "function"];
1306
+ readonly jsonOutput: true;
1307
+ readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "response_format", "tools", "tool_choice"];
1235
1308
  }];
1236
1309
  }, {
1237
1310
  readonly id: "qwen3-vl-30b-a3b-thinking";
@@ -1874,6 +1947,7 @@ export declare const alibabaModels: [{
1874
1947
  readonly vision: true;
1875
1948
  readonly tools: true;
1876
1949
  readonly jsonOutput: true;
1950
+ readonly deactivatedAt: Date;
1877
1951
  }];
1878
1952
  }, {
1879
1953
  readonly id: "qwen3-vl-235b-a22b-instruct";
@@ -1906,6 +1980,21 @@ export declare const alibabaModels: [{
1906
1980
  readonly vision: true;
1907
1981
  readonly tools: true;
1908
1982
  readonly jsonOutput: true;
1983
+ }, {
1984
+ readonly providerId: "deepinfra";
1985
+ readonly externalId: "Qwen/Qwen3-VL-235B-A22B-Instruct";
1986
+ readonly inputPrice: "0.20e-6";
1987
+ readonly outputPrice: "0.88e-6";
1988
+ readonly cachedInputPrice: "0.11e-6";
1989
+ readonly requestPrice: "0";
1990
+ readonly contextSize: 262144;
1991
+ readonly maxOutput: 32768;
1992
+ readonly quantization: "fp8";
1993
+ readonly streaming: true;
1994
+ readonly vision: true;
1995
+ readonly tools: true;
1996
+ readonly jsonOutput: true;
1997
+ readonly supportedParameters: ["temperature", "max_tokens", "top_p", "stream", "tools", "response_format"];
1909
1998
  }];
1910
1999
  }, {
1911
2000
  readonly id: "qwen3-vl-235b-a22b-thinking";
@@ -2151,6 +2240,7 @@ export declare const alibabaModels: [{
2151
2240
  readonly tools: true;
2152
2241
  readonly jsonOutput: true;
2153
2242
  readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "stream", "response_format", "tools", "reasoning_effort"];
2243
+ readonly deactivatedAt: Date;
2154
2244
  }];
2155
2245
  }, {
2156
2246
  readonly id: "qwen3.6-plus";
@@ -2224,29 +2314,92 @@ export declare const alibabaModels: [{
2224
2314
  }, {
2225
2315
  readonly id: "qwen3.6-flash";
2226
2316
  readonly name: "Qwen3.6 Flash";
2227
- readonly description: "Fast, cost-effective Qwen3.6 model with hybrid reasoning, tool calling, and a 262K context window for high-volume tasks.";
2317
+ readonly description: "Fast, cost-effective Qwen3.6 model with native vision-language capabilities, hybrid reasoning, tool calling, and a 1M context window for high-volume tasks.";
2228
2318
  readonly family: "alibaba";
2229
2319
  readonly releasedAt: Date;
2230
2320
  readonly providers: [{
2231
2321
  readonly providerId: "alibaba";
2232
2322
  readonly externalId: "qwen3.6-flash";
2233
- readonly inputPrice: "0.15e-6";
2234
- readonly outputPrice: "0.9e-6";
2235
- readonly cachedInputPrice: "0.015e-6";
2323
+ readonly inputPrice: "0.25e-6";
2324
+ readonly outputPrice: "1.5e-6";
2325
+ readonly cachedInputPrice: "0.05e-6";
2326
+ readonly cacheReadInputPrice: "0.025e-6";
2327
+ readonly cacheWriteInputPrice: "0.3125e-6";
2328
+ readonly pricingTiers: [{
2329
+ readonly name: "Up to 256K";
2330
+ readonly upToTokens: 256000;
2331
+ readonly inputPrice: "0.25e-6";
2332
+ readonly outputPrice: "1.5e-6";
2333
+ readonly cachedInputPrice: "0.05e-6";
2334
+ readonly cacheReadInputPrice: "0.025e-6";
2335
+ readonly cacheWriteInputPrice: "0.3125e-6";
2336
+ }, {
2337
+ readonly name: "Over 256K";
2338
+ readonly upToTokens: number;
2339
+ readonly inputPrice: "1.0e-6";
2340
+ readonly outputPrice: "4e-6";
2341
+ readonly cachedInputPrice: "0.2e-6";
2342
+ readonly cacheReadInputPrice: "0.1e-6";
2343
+ readonly cacheWriteInputPrice: "1.25e-6";
2344
+ }];
2236
2345
  readonly regions: [{
2237
2346
  readonly id: "singapore";
2238
2347
  }, {
2239
2348
  readonly id: "us-virginia";
2349
+ readonly inputPrice: "0.165e-6";
2350
+ readonly outputPrice: "0.99e-6";
2351
+ readonly cachedInputPrice: "0.033e-6";
2352
+ readonly cacheReadInputPrice: "0.0165e-6";
2353
+ readonly cacheWriteInputPrice: "0.20625e-6";
2354
+ readonly pricingTiers: [{
2355
+ readonly name: "Up to 256K";
2356
+ readonly upToTokens: 256000;
2357
+ readonly inputPrice: "0.165e-6";
2358
+ readonly outputPrice: "0.99e-6";
2359
+ readonly cachedInputPrice: "0.033e-6";
2360
+ readonly cacheReadInputPrice: "0.0165e-6";
2361
+ readonly cacheWriteInputPrice: "0.20625e-6";
2362
+ }, {
2363
+ readonly name: "Over 256K";
2364
+ readonly upToTokens: number;
2365
+ readonly inputPrice: "0.66e-6";
2366
+ readonly outputPrice: "3.961e-6";
2367
+ readonly cachedInputPrice: "0.132e-6";
2368
+ readonly cacheReadInputPrice: "0.066e-6";
2369
+ readonly cacheWriteInputPrice: "0.825e-6";
2370
+ }];
2240
2371
  }, {
2241
2372
  readonly id: "cn-beijing";
2373
+ readonly inputPrice: "0.165e-6";
2374
+ readonly outputPrice: "0.99e-6";
2375
+ readonly cachedInputPrice: "0.033e-6";
2376
+ readonly cacheReadInputPrice: "0.0165e-6";
2377
+ readonly cacheWriteInputPrice: "0.20625e-6";
2378
+ readonly pricingTiers: [{
2379
+ readonly name: "Up to 256K";
2380
+ readonly upToTokens: 256000;
2381
+ readonly inputPrice: "0.165e-6";
2382
+ readonly outputPrice: "0.99e-6";
2383
+ readonly cachedInputPrice: "0.033e-6";
2384
+ readonly cacheReadInputPrice: "0.0165e-6";
2385
+ readonly cacheWriteInputPrice: "0.20625e-6";
2386
+ }, {
2387
+ readonly name: "Over 256K";
2388
+ readonly upToTokens: number;
2389
+ readonly inputPrice: "0.66e-6";
2390
+ readonly outputPrice: "3.961e-6";
2391
+ readonly cachedInputPrice: "0.132e-6";
2392
+ readonly cacheReadInputPrice: "0.066e-6";
2393
+ readonly cacheWriteInputPrice: "0.825e-6";
2394
+ }];
2242
2395
  }];
2243
2396
  readonly requestPrice: "0";
2244
- readonly contextSize: 262144;
2397
+ readonly contextSize: 1000000;
2245
2398
  readonly maxOutput: 65536;
2246
2399
  readonly reasoning: true;
2247
2400
  readonly reasoningMaxTokens: true;
2248
2401
  readonly streaming: true;
2249
- readonly vision: false;
2402
+ readonly vision: true;
2250
2403
  readonly tools: true;
2251
2404
  readonly jsonOutput: true;
2252
2405
  readonly supportedParameters: ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "stream", "response_format", "tools", "reasoning_effort"];
package/dist/models/alibaba.js CHANGED
@@ -430,6 +430,7 @@ export const alibabaModels = [
430
430
  vision: false,
431
431
  tools: true,
432
432
  jsonOutput: true,
433
+ deactivatedAt: new Date("2026-10-10"),
433
434
  },
434
435
  ],
435
436
  },
@@ -878,7 +879,7 @@ export const alibabaModels = [
878
879
  {
879
880
  providerId: "novita",
880
881
  externalId: "qwen/qwen3-vl-8b-instruct",
881
- deactivatedAt: new Date("2026-07-11"),
882
+ deactivatedAt: new Date("2026-07-01"),
882
883
  inputPrice: "0.08e-6",
883
884
  outputPrice: "0.5e-6",
884
885
  requestPrice: "0",
@@ -1146,7 +1147,7 @@ export const alibabaModels = [
1146
1147
  {
1147
1148
  providerId: "novita",
1148
1149
  externalId: "qwen/qwen3-next-80b-a3b-thinking",
1149
- deactivatedAt: new Date("2026-07-11"),
1150
+ deactivatedAt: new Date("2026-07-01"),
1150
1151
  inputPrice: "0.15e-6",
1151
1152
  outputPrice: "1.5e-6",
1152
1153
  requestPrice: "0",
@@ -1302,6 +1303,7 @@ export const alibabaModels = [
1302
1303
  vision: true,
1303
1304
  tools: true,
1304
1305
  jsonOutput: true,
1306
+ deactivatedAt: new Date("2026-10-10"),
1305
1307
  },
1306
1308
  {
1307
1309
  providerId: "novita",
@@ -1495,6 +1497,76 @@ export const alibabaModels = [
1495
1497
  },
1496
1498
  ],
1497
1499
  },
1500
+ {
1501
+ id: "qwen3.7-flash",
1502
+ name: "Qwen3.7 Flash",
1503
+ description: "Fast, cost-effective Qwen3.7 model with hybrid reasoning and tool calling for high-volume tasks.",
1504
+ family: "alibaba",
1505
+ releasedAt: new Date("2026-07-27"),
1506
+ providers: [
1507
+ {
1508
+ providerId: "alibaba",
1509
+ externalId: "qwen3.7-flash",
1510
+ inputPrice: "0.03e-6",
1511
+ outputPrice: "0.13e-6",
1512
+ cachedInputPrice: "0.006e-6",
1513
+ cacheReadInputPrice: "0.003e-6",
1514
+ cacheWriteInputPrice: "0.0375e-6",
1515
+ pricingTiers: [
1516
+ {
1517
+ name: "Up to 32K",
1518
+ upToTokens: 32000,
1519
+ inputPrice: "0.03e-6",
1520
+ outputPrice: "0.13e-6",
1521
+ cachedInputPrice: "0.006e-6",
1522
+ cacheReadInputPrice: "0.003e-6",
1523
+ cacheWriteInputPrice: "0.0375e-6",
1524
+ },
1525
+ {
1526
+ name: "32K-256K",
1527
+ upToTokens: 256000,
1528
+ inputPrice: "0.10e-6",
1529
+ outputPrice: "0.40e-6",
1530
+ cachedInputPrice: "0.02e-6",
1531
+ cacheReadInputPrice: "0.01e-6",
1532
+ cacheWriteInputPrice: "0.125e-6",
1533
+ },
1534
+ {
1535
+ name: "256K-1M",
1536
+ upToTokens: 1000000,
1537
+ inputPrice: "0.20e-6",
1538
+ outputPrice: "0.80e-6",
1539
+ cachedInputPrice: "0.04e-6",
1540
+ cacheReadInputPrice: "0.02e-6",
1541
+ cacheWriteInputPrice: "0.25e-6",
1542
+ },
1543
+ ],
1544
+ regions: [{ id: "singapore" }, { id: "cn-beijing" }],
1545
+ requestPrice: "0",
1546
+ contextSize: 1000000,
1547
+ maxOutput: 65536,
1548
+ reasoning: true,
1549
+ reasoningMaxTokens: true,
1550
+ reasoningOutput: "omit",
1551
+ streaming: true,
1552
+ vision: true,
1553
+ tools: true,
1554
+ jsonOutput: true,
1555
+ supportedParameters: [
1556
+ "temperature",
1557
+ "max_tokens",
1558
+ "top_p",
1559
+ "frequency_penalty",
1560
+ "presence_penalty",
1561
+ "stop",
1562
+ "stream",
1563
+ "tools",
1564
+ "response_format",
1565
+ "reasoning_effort",
1566
+ ],
1567
+ },
1568
+ ],
1569
+ },
1498
1570
  {
1499
1571
  id: "qwen3-coder-next",
1500
1572
  name: "Qwen3 Coder Next",
@@ -1563,6 +1635,30 @@ export const alibabaModels = [
1563
1635
  "tool_choice",
1564
1636
  ],
1565
1637
  },
1638
+ {
1639
+ providerId: "deepinfra",
1640
+ externalId: "Qwen/Qwen3-VL-30B-A3B-Instruct",
1641
+ inputPrice: "0.15e-6",
1642
+ outputPrice: "0.60e-6",
1643
+ requestPrice: "0",
1644
+ contextSize: 262144,
1645
+ maxOutput: 32768,
1646
+ quantization: "fp8",
1647
+ streaming: true,
1648
+ vision: true,
1649
+ tools: true,
1650
+ supportedToolChoices: ["auto", "none", "function"],
1651
+ jsonOutput: true,
1652
+ supportedParameters: [
1653
+ "temperature",
1654
+ "max_tokens",
1655
+ "top_p",
1656
+ "frequency_penalty",
1657
+ "response_format",
1658
+ "tools",
1659
+ "tool_choice",
1660
+ ],
1661
+ },
1566
1662
  ],
1567
1663
  },
1568
1664
  {
@@ -1575,7 +1671,7 @@ export const alibabaModels = [
1575
1671
  {
1576
1672
  providerId: "novita",
1577
1673
  externalId: "qwen/qwen3-vl-30b-a3b-thinking",
1578
- deactivatedAt: new Date("2026-07-11"),
1674
+ deactivatedAt: new Date("2026-07-01"),
1579
1675
  inputPrice: "0.2e-6",
1580
1676
  outputPrice: "1e-6",
1581
1677
  requestPrice: "0",
@@ -2333,6 +2429,7 @@ export const alibabaModels = [
2333
2429
  vision: true,
2334
2430
  tools: true,
2335
2431
  jsonOutput: true,
2432
+ deactivatedAt: new Date("2026-10-10"),
2336
2433
  },
2337
2434
  ],
2338
2435
  },
@@ -2370,6 +2467,29 @@ export const alibabaModels = [
2370
2467
  tools: true,
2371
2468
  jsonOutput: true,
2372
2469
  },
2470
+ {
2471
+ providerId: "deepinfra",
2472
+ externalId: "Qwen/Qwen3-VL-235B-A22B-Instruct",
2473
+ inputPrice: "0.20e-6",
2474
+ outputPrice: "0.88e-6",
2475
+ cachedInputPrice: "0.11e-6",
2476
+ requestPrice: "0",
2477
+ contextSize: 262144,
2478
+ maxOutput: 32768,
2479
+ quantization: "fp8",
2480
+ streaming: true,
2481
+ vision: true,
2482
+ tools: true,
2483
+ jsonOutput: true,
2484
+ supportedParameters: [
2485
+ "temperature",
2486
+ "max_tokens",
2487
+ "top_p",
2488
+ "stream",
2489
+ "tools",
2490
+ "response_format",
2491
+ ],
2492
+ },
2373
2493
  ],
2374
2494
  },
2375
2495
  {
@@ -2658,6 +2778,7 @@ export const alibabaModels = [
2658
2778
  "tools",
2659
2779
  "reasoning_effort",
2660
2780
  ],
2781
+ deactivatedAt: new Date("2026-10-10"),
2661
2782
  },
2662
2783
  ],
2663
2784
  },
@@ -2768,28 +2889,104 @@ export const alibabaModels = [
2768
2889
  {
2769
2890
  id: "qwen3.6-flash",
2770
2891
  name: "Qwen3.6 Flash",
2771
- description: "Fast, cost-effective Qwen3.6 model with hybrid reasoning, tool calling, and a 262K context window for high-volume tasks.",
2892
+ description: "Fast, cost-effective Qwen3.6 model with native vision-language capabilities, hybrid reasoning, tool calling, and a 1M context window for high-volume tasks.",
2772
2893
  family: "alibaba",
2773
2894
  releasedAt: new Date("2026-04-29"),
2774
2895
  providers: [
2775
2896
  {
2776
2897
  providerId: "alibaba",
2777
2898
  externalId: "qwen3.6-flash",
2778
- inputPrice: "0.15e-6",
2779
- outputPrice: "0.9e-6",
2780
- cachedInputPrice: "0.015e-6",
2899
+ inputPrice: "0.25e-6",
2900
+ outputPrice: "1.5e-6",
2901
+ cachedInputPrice: "0.05e-6",
2902
+ cacheReadInputPrice: "0.025e-6",
2903
+ cacheWriteInputPrice: "0.3125e-6",
2904
+ pricingTiers: [
2905
+ {
2906
+ name: "Up to 256K",
2907
+ upToTokens: 256000,
2908
+ inputPrice: "0.25e-6",
2909
+ outputPrice: "1.5e-6",
2910
+ cachedInputPrice: "0.05e-6",
2911
+ cacheReadInputPrice: "0.025e-6",
2912
+ cacheWriteInputPrice: "0.3125e-6",
2913
+ },
2914
+ {
2915
+ name: "Over 256K",
2916
+ upToTokens: Infinity,
2917
+ inputPrice: "1.0e-6",
2918
+ outputPrice: "4e-6",
2919
+ cachedInputPrice: "0.2e-6",
2920
+ cacheReadInputPrice: "0.1e-6",
2921
+ cacheWriteInputPrice: "1.25e-6",
2922
+ },
2923
+ ],
2781
2924
  regions: [
2782
2925
  { id: "singapore" },
2783
- { id: "us-virginia" },
2784
- { id: "cn-beijing" },
2926
+ {
2927
+ id: "us-virginia",
2928
+ inputPrice: "0.165e-6",
2929
+ outputPrice: "0.99e-6",
2930
+ cachedInputPrice: "0.033e-6",
2931
+ cacheReadInputPrice: "0.0165e-6",
2932
+ cacheWriteInputPrice: "0.20625e-6",
2933
+ pricingTiers: [
2934
+ {
2935
+ name: "Up to 256K",
2936
+ upToTokens: 256000,
2937
+ inputPrice: "0.165e-6",
2938
+ outputPrice: "0.99e-6",
2939
+ cachedInputPrice: "0.033e-6",
2940
+ cacheReadInputPrice: "0.0165e-6",
2941
+ cacheWriteInputPrice: "0.20625e-6",
2942
+ },
2943
+ {
2944
+ name: "Over 256K",
2945
+ upToTokens: Infinity,
2946
+ inputPrice: "0.66e-6",
2947
+ outputPrice: "3.961e-6",
2948
+ cachedInputPrice: "0.132e-6",
2949
+ cacheReadInputPrice: "0.066e-6",
2950
+ cacheWriteInputPrice: "0.825e-6",
2951
+ },
2952
+ ],
2953
+ },
2954
+ {
2955
+ id: "cn-beijing",
2956
+ inputPrice: "0.165e-6",
2957
+ outputPrice: "0.99e-6",
2958
+ cachedInputPrice: "0.033e-6",
2959
+ cacheReadInputPrice: "0.0165e-6",
2960
+ cacheWriteInputPrice: "0.20625e-6",
2961
+ pricingTiers: [
2962
+ {
2963
+ name: "Up to 256K",
2964
+ upToTokens: 256000,
2965
+ inputPrice: "0.165e-6",
2966
+ outputPrice: "0.99e-6",
2967
+ cachedInputPrice: "0.033e-6",
2968
+ cacheReadInputPrice: "0.0165e-6",
2969
+ cacheWriteInputPrice: "0.20625e-6",
2970
+ },
2971
+ {
2972
+ name: "Over 256K",
2973
+ upToTokens: Infinity,
2974
+ inputPrice: "0.66e-6",
2975
+ outputPrice: "3.961e-6",
2976
+ cachedInputPrice: "0.132e-6",
2977
+ cacheReadInputPrice: "0.066e-6",
2978
+ cacheWriteInputPrice: "0.825e-6",
2979
+ },
2980
+ ],
2981
+ },
2785
2982
  ],
2786
2983
  requestPrice: "0",
2787
- contextSize: 262144,
2984
+ contextSize: 1000000,
2788
2985
  maxOutput: 65536,
2789
2986
  reasoning: true,
2790
2987
  reasoningMaxTokens: true,
2791
2988
  streaming: true,
2792
- vision: false,
2989
+ vision: true,
2793
2990
  tools: true,
2794
2991
  jsonOutput: true,
2795
2992
  supportedParameters: [