@llmgateway/models 1.303.0 → 1.305.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/models.d.ts CHANGED
@@ -99,6 +99,7 @@ export interface ProviderModelMapping {
99
99
  supportsNStreaming?: boolean;
100
100
  maxN?: number;
101
101
  reasoningOutput?: "omit";
102
+ reasoningSummary?: "auto" | "detailed";
102
103
  reasoningMaxTokens?: boolean;
103
104
  reasoningMode?: "enabled" | "adaptive";
104
105
  reasoningEfforts?: ReasoningEffort[];
@@ -2149,6 +2150,89 @@ export declare const models: [{
2149
2150
  readonly jsonOutputSchema: true;
2150
2151
  readonly supportedParameters: ["max_tokens", "response_format", "verbosity", "tools", "tool_choice"];
2151
2152
  readonly jsonOutput: true;
2153
+ }, {
2154
+ readonly providerId: "aws-mantle";
2155
+ readonly externalId: "openai.gpt-6-astra";
2156
+ readonly reasoningSummary: "auto";
2157
+ readonly regions: [{
2158
+ readonly id: "global";
2159
+ }, {
2160
+ readonly id: "us";
2161
+ readonly inputPrice: "11.0e-6";
2162
+ readonly outputPrice: "55.0e-6";
2163
+ readonly cachedInputPrice: "1.1e-6";
2164
+ readonly cacheWriteInputPrice: "13.75e-6";
2165
+ readonly pricingTiers: [{
2166
+ readonly name: "Up to 272K";
2167
+ readonly upToTokens: 272000;
2168
+ readonly inputPrice: "11.0e-6";
2169
+ readonly outputPrice: "55.0e-6";
2170
+ readonly cachedInputPrice: "1.1e-6";
2171
+ readonly cacheWriteInputPrice: "13.75e-6";
2172
+ }, {
2173
+ readonly name: "Over 272K";
2174
+ readonly upToTokens: number;
2175
+ readonly inputPrice: "22.0e-6";
2176
+ readonly outputPrice: "82.5e-6";
2177
+ readonly cachedInputPrice: "2.2e-6";
2178
+ readonly cacheWriteInputPrice: "27.5e-6";
2179
+ }];
2180
+ }, {
2181
+ readonly id: "us-west-2";
2182
+ readonly inputPrice: "11.0e-6";
2183
+ readonly outputPrice: "55.0e-6";
2184
+ readonly cachedInputPrice: "1.1e-6";
2185
+ readonly cacheWriteInputPrice: "13.75e-6";
2186
+ readonly pricingTiers: [{
2187
+ readonly name: "Up to 272K";
2188
+ readonly upToTokens: 272000;
2189
+ readonly inputPrice: "11.0e-6";
2190
+ readonly outputPrice: "55.0e-6";
2191
+ readonly cachedInputPrice: "1.1e-6";
2192
+ readonly cacheWriteInputPrice: "13.75e-6";
2193
+ }, {
2194
+ readonly name: "Over 272K";
2195
+ readonly upToTokens: number;
2196
+ readonly inputPrice: "22.0e-6";
2197
+ readonly outputPrice: "82.5e-6";
2198
+ readonly cachedInputPrice: "2.2e-6";
2199
+ readonly cacheWriteInputPrice: "27.5e-6";
2200
+ }];
2201
+ }];
2202
+ readonly inputPrice: "10.0e-6";
2203
+ readonly outputPrice: "50.0e-6";
2204
+ readonly cachedInputPrice: "1.0e-6";
2205
+ readonly cacheWriteInputPrice: "12.5e-6";
2206
+ readonly pricingTiers: [{
2207
+ readonly name: "Up to 272K";
2208
+ readonly upToTokens: 272000;
2209
+ readonly inputPrice: "10.0e-6";
2210
+ readonly outputPrice: "50.0e-6";
2211
+ readonly cachedInputPrice: "1.0e-6";
2212
+ readonly cacheWriteInputPrice: "12.5e-6";
2213
+ }, {
2214
+ readonly name: "Over 272K";
2215
+ readonly upToTokens: number;
2216
+ readonly inputPrice: "20.0e-6";
2217
+ readonly outputPrice: "75.0e-6";
2218
+ readonly cachedInputPrice: "2.0e-6";
2219
+ readonly cacheWriteInputPrice: "25.0e-6";
2220
+ }];
2221
+ readonly requestPrice: "0";
2222
+ readonly contextSize: 1050000;
2223
+ readonly maxOutput: 128000;
2224
+ readonly streaming: true;
2225
+ readonly vision: true;
2226
+ readonly tools: true;
2227
+ readonly supportedToolChoices: ["auto", "none", "required", "function"];
2228
+ readonly reasoning: true;
2229
+ readonly reasoningEfforts: ["low", "medium", "high", "xhigh", "max"];
2230
+ readonly reasoningOutput: "omit";
2231
+ readonly verbosity: true;
2232
+ readonly supportsResponsesApi: true;
2233
+ readonly jsonOutputSchema: true;
2234
+ readonly supportedParameters: ["max_tokens", "response_format", "verbosity", "tools", "tool_choice"];
2235
+ readonly jsonOutput: true;
2152
2236
  }];
2153
2237
  }, {
2154
2238
  readonly id: "gpt-5.2-codex";
@@ -8369,6 +8453,22 @@ export declare const models: [{
8369
8453
  readonly vision: false;
8370
8454
  readonly tools: false;
8371
8455
  readonly jsonOutput: false;
8456
+ }, {
8457
+ readonly providerId: "tencent";
8458
+ readonly externalId: "mimo-v2.5-pro";
8459
+ readonly inputPrice: "0.435e-6";
8460
+ readonly cachedInputPrice: "0.0036e-6";
8461
+ readonly outputPrice: "0.87e-6";
8462
+ readonly requestPrice: "0";
8463
+ readonly contextSize: 1000000;
8464
+ readonly maxOutput: 131072;
8465
+ readonly streaming: true;
8466
+ readonly reasoning: true;
8467
+ readonly reasoningEfforts: ["none", "low", "medium", "high"];
8468
+ readonly vision: false;
8469
+ readonly tools: true;
8470
+ readonly jsonOutput: true;
8471
+ readonly jsonOutputSchema: false;
8372
8472
  }];
8373
8473
  }, {
8374
8474
  readonly id: "mimo-v2-pro";
@@ -9504,6 +9604,23 @@ export declare const models: [{
9504
9604
  readonly vision: false;
9505
9605
  readonly tools: true;
9506
9606
  readonly jsonOutput: true;
9607
+ }, {
9608
+ readonly providerId: "tencent";
9609
+ readonly externalId: "deepseek-v4-pro-202606";
9610
+ readonly inputPrice: "0.435e-6";
9611
+ readonly cachedInputPrice: "0.00363e-6";
9612
+ readonly outputPrice: "0.87e-6";
9613
+ readonly requestPrice: "0";
9614
+ readonly contextSize: 1000000;
9615
+ readonly maxOutput: 393216;
9616
+ readonly streaming: true;
9617
+ readonly reasoning: true;
9618
+ readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
9619
+ readonly vision: false;
9620
+ readonly tools: true;
9621
+ readonly supportedToolChoices: ["auto", "none"];
9622
+ readonly jsonOutput: true;
9623
+ readonly jsonOutputSchema: false;
9507
9624
  }];
9508
9625
  }, {
9509
9626
  readonly id: "deepseek-v4-flash";
@@ -9762,6 +9879,23 @@ export declare const models: [{
9762
9879
  readonly supportsAssistantPrefill: false;
9763
9880
  readonly jsonOutput: true;
9764
9881
  readonly jsonOutputSchema: true;
9882
+ }, {
9883
+ readonly providerId: "tencent";
9884
+ readonly externalId: "deepseek-v4-flash-202605";
9885
+ readonly inputPrice: "0.14e-6";
9886
+ readonly cachedInputPrice: "0.0028e-6";
9887
+ readonly outputPrice: "0.28e-6";
9888
+ readonly requestPrice: "0";
9889
+ readonly contextSize: 1000000;
9890
+ readonly maxOutput: 393216;
9891
+ readonly streaming: true;
9892
+ readonly reasoning: true;
9893
+ readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
9894
+ readonly vision: false;
9895
+ readonly tools: true;
9896
+ readonly supportedToolChoices: ["auto", "none"];
9897
+ readonly jsonOutput: true;
9898
+ readonly jsonOutputSchema: false;
9765
9899
  }];
9766
9900
  }, {
9767
9901
  readonly id: "deepseek-v4-flash-vision-exp";
@@ -10120,6 +10254,35 @@ export declare const models: [{
10120
10254
  readonly vision: false;
10121
10255
  readonly tools: true;
10122
10256
  readonly jsonOutput: false;
10257
+ }, {
10258
+ readonly providerId: "tencent";
10259
+ readonly externalId: "minimax-m3";
10260
+ readonly inputPrice: "0.3e-6";
10261
+ readonly cachedInputPrice: "0.06e-6";
10262
+ readonly outputPrice: "1.2e-6";
10263
+ readonly pricingTiers: [{
10264
+ readonly name: "Up to 512K";
10265
+ readonly upToTokens: 512000;
10266
+ readonly inputPrice: "0.3e-6";
10267
+ readonly outputPrice: "1.2e-6";
10268
+ readonly cachedInputPrice: "0.06e-6";
10269
+ }, {
10270
+ readonly name: "Over 512K";
10271
+ readonly upToTokens: number;
10272
+ readonly inputPrice: "0.6e-6";
10273
+ readonly outputPrice: "2.4e-6";
10274
+ readonly cachedInputPrice: "0.12e-6";
10275
+ }];
10276
+ readonly requestPrice: "0";
10277
+ readonly contextSize: 1000000;
10278
+ readonly maxOutput: 131072;
10279
+ readonly streaming: true;
10280
+ readonly reasoning: true;
10281
+ readonly reasoningEfforts: ["low", "medium", "high", "xhigh", "max"];
10282
+ readonly splitTaggedReasoning: true;
10283
+ readonly vision: true;
10284
+ readonly tools: true;
10285
+ readonly jsonOutput: false;
10123
10286
  }];
10124
10287
  }, {
10125
10288
  readonly id: "minimax-m2.7";
@@ -10201,6 +10364,22 @@ export declare const models: [{
10201
10364
  readonly vision: false;
10202
10365
  readonly tools: true;
10203
10366
  readonly jsonOutput: true;
10367
+ }, {
10368
+ readonly providerId: "tencent";
10369
+ readonly externalId: "minimax-m2.7";
10370
+ readonly inputPrice: "0.3e-6";
10371
+ readonly cachedInputPrice: "0.06e-6";
10372
+ readonly outputPrice: "1.2e-6";
10373
+ readonly requestPrice: "0";
10374
+ readonly contextSize: 200000;
10375
+ readonly maxOutput: 128000;
10376
+ readonly streaming: true;
10377
+ readonly reasoning: true;
10378
+ readonly reasoningEfforts: ["low", "medium", "high", "xhigh", "max"];
10379
+ readonly splitTaggedReasoning: true;
10380
+ readonly vision: false;
10381
+ readonly tools: true;
10382
+ readonly jsonOutput: false;
10204
10383
  }];
10205
10384
  }, {
10206
10385
  readonly id: "minimax-m2.7-highspeed";
@@ -10907,6 +11086,23 @@ export declare const models: [{
10907
11086
  readonly vision: true;
10908
11087
  readonly tools: true;
10909
11088
  readonly jsonOutput: true;
11089
+ }, {
11090
+ readonly providerId: "tencent";
11091
+ readonly externalId: "kimi-k2.6";
11092
+ readonly inputPrice: "0.858e-6";
11093
+ readonly cachedInputPrice: "0.145e-6";
11094
+ readonly outputPrice: "3.566e-6";
11095
+ readonly requestPrice: "0";
11096
+ readonly contextSize: 262144;
11097
+ readonly maxOutput: 262144;
11098
+ readonly streaming: true;
11099
+ readonly reasoning: true;
11100
+ readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
11101
+ readonly vision: true;
11102
+ readonly tools: true;
11103
+ readonly supportedToolChoices: ["auto", "none"];
11104
+ readonly jsonOutput: true;
11105
+ readonly jsonOutputSchema: true;
10910
11106
  }];
10911
11107
  }, {
10912
11108
  readonly id: "kimi-k2.7-code";
@@ -10982,6 +11178,23 @@ export declare const models: [{
10982
11178
  readonly tools: true;
10983
11179
  readonly jsonOutput: true;
10984
11180
  readonly supportedParameters: ["max_tokens", "response_format", "tools", "tool_choice", "reasoning_effort"];
11181
+ }, {
11182
+ readonly providerId: "tencent";
11183
+ readonly externalId: "kimi-k2.7-code";
11184
+ readonly inputPrice: "0.95e-6";
11185
+ readonly cachedInputPrice: "0.19e-6";
11186
+ readonly outputPrice: "4.0e-6";
11187
+ readonly requestPrice: "0";
11188
+ readonly contextSize: 262144;
11189
+ readonly maxOutput: 262144;
11190
+ readonly streaming: true;
11191
+ readonly reasoning: true;
11192
+ readonly reasoningEfforts: ["minimal", "low", "medium", "high", "xhigh", "max"];
11193
+ readonly vision: true;
11194
+ readonly tools: true;
11195
+ readonly supportedToolChoices: ["auto", "none"];
11196
+ readonly jsonOutput: true;
11197
+ readonly jsonOutputSchema: true;
10985
11198
  }];
10986
11199
  }, {
10987
11200
  readonly id: "kimi-k2.7-code-highspeed";
@@ -11006,6 +11219,23 @@ export declare const models: [{
11006
11219
  readonly tools: true;
11007
11220
  readonly jsonOutput: false;
11008
11221
  readonly supportedParameters: ["max_tokens", "tools", "tool_choice", "reasoning_effort"];
11222
+ }, {
11223
+ readonly providerId: "tencent";
11224
+ readonly externalId: "kimi-k2.7-code-highspeed";
11225
+ readonly inputPrice: "1.9e-6";
11226
+ readonly cachedInputPrice: "0.38e-6";
11227
+ readonly outputPrice: "8e-6";
11228
+ readonly requestPrice: "0";
11229
+ readonly contextSize: 262144;
11230
+ readonly maxOutput: 262144;
11231
+ readonly streaming: true;
11232
+ readonly reasoning: true;
11233
+ readonly reasoningEfforts: ["minimal", "low", "medium", "high", "xhigh", "max"];
11234
+ readonly vision: true;
11235
+ readonly tools: true;
11236
+ readonly supportedToolChoices: ["auto", "none"];
11237
+ readonly jsonOutput: true;
11238
+ readonly jsonOutputSchema: true;
11009
11239
  }];
11010
11240
  }, {
11011
11241
  readonly id: "kimi-k3";
@@ -11208,6 +11438,23 @@ export declare const models: [{
11208
11438
  readonly jsonOutput: true;
11209
11439
  readonly jsonOutputSchema: true;
11210
11440
  readonly supportedParameters: ["max_tokens", "response_format", "tools", "tool_choice", "reasoning_effort"];
11441
+ }, {
11442
+ readonly providerId: "tencent";
11443
+ readonly externalId: "kimi-k3";
11444
+ readonly inputPrice: "3.0e-6";
11445
+ readonly cachedInputPrice: "0.3e-6";
11446
+ readonly outputPrice: "15.0e-6";
11447
+ readonly requestPrice: "0";
11448
+ readonly contextSize: 1048576;
11449
+ readonly maxOutput: 1048576;
11450
+ readonly streaming: true;
11451
+ readonly reasoning: true;
11452
+ readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
11453
+ readonly vision: true;
11454
+ readonly tools: true;
11455
+ readonly supportedToolChoices: ["auto", "none", "required"];
11456
+ readonly jsonOutput: true;
11457
+ readonly jsonOutputSchema: true;
11211
11458
  }];
11212
11459
  }, {
11213
11460
  readonly id: "kimi-k3-fast";
@@ -15212,6 +15459,44 @@ export declare const models: [{
15212
15459
  readonly jsonOutput: true;
15213
15460
  readonly jsonOutputSchema: true;
15214
15461
  }];
15462
+ }, {
15463
+ readonly id: "kinfra-text-embedding-0.6b";
15464
+ readonly name: "Kinfra Text Embedding 0.6B";
15465
+ readonly description: "Lightweight text embedding model with 1024-dimensional output and a 32K context, aimed at large-scale retrieval where latency and cost matter most.";
15466
+ readonly family: "kinfra";
15467
+ readonly output: ["embedding"];
15468
+ readonly releasedAt: Date;
15469
+ readonly providers: [{
15470
+ readonly providerId: "tencent";
15471
+ readonly externalId: "kinfra-text-embedding-0.6b";
15472
+ readonly inputPrice: "0.07e-6";
15473
+ readonly outputPrice: "0";
15474
+ readonly requestPrice: "0";
15475
+ readonly contextSize: 32768;
15476
+ readonly streaming: false;
15477
+ readonly tools: false;
15478
+ readonly jsonOutput: false;
15479
+ readonly embeddings: true;
15480
+ }];
15481
+ }, {
15482
+ readonly id: "kinfra-text-embedding-4b";
15483
+ readonly name: "Kinfra Text Embedding 4B";
15484
+ readonly description: "High-quality text embedding model with 2560-dimensional output and a 32K context, aimed at deep semantic search.";
15485
+ readonly family: "kinfra";
15486
+ readonly output: ["embedding"];
15487
+ readonly releasedAt: Date;
15488
+ readonly providers: [{
15489
+ readonly providerId: "tencent";
15490
+ readonly externalId: "kinfra-text-embedding-4b";
15491
+ readonly inputPrice: "0.084e-6";
15492
+ readonly outputPrice: "0";
15493
+ readonly requestPrice: "0";
15494
+ readonly contextSize: 32768;
15495
+ readonly streaming: false;
15496
+ readonly tools: false;
15497
+ readonly jsonOutput: false;
15498
+ readonly embeddings: true;
15499
+ }];
15215
15500
  }, {
15216
15501
  readonly id: "hy3";
15217
15502
  readonly name: "Hy3";
@@ -15252,6 +15537,65 @@ export declare const models: [{
15252
15537
  readonly tools: true;
15253
15538
  readonly supportedToolChoices: ["auto", "none", "required"];
15254
15539
  readonly jsonOutputSchema: true;
15540
+ }, {
15541
+ readonly providerId: "tencent";
15542
+ readonly externalId: "hy3";
15543
+ readonly inputPrice: "0.132e-6";
15544
+ readonly cachedInputPrice: "0.033e-6";
15545
+ readonly outputPrice: "0.528e-6";
15546
+ readonly requestPrice: "0";
15547
+ readonly contextSize: 262144;
15548
+ readonly maxOutput: 131072;
15549
+ readonly streaming: true;
15550
+ readonly reasoning: true;
15551
+ readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh"];
15552
+ readonly vision: false;
15553
+ readonly tools: true;
15554
+ readonly jsonOutput: true;
15555
+ readonly jsonOutputSchema: true;
15556
+ }];
15557
+ }, {
15558
+ readonly id: "hy-mt2-plus";
15559
+ readonly name: "Hy-MT2 Plus";
15560
+ readonly description: "Tencent's Hunyuan translation model, tuned for translation quality and instruction following across a short 8K context.";
15561
+ readonly family: "tencent";
15562
+ readonly releasedAt: Date;
15563
+ readonly providers: [{
15564
+ readonly providerId: "tencent";
15565
+ readonly externalId: "hy-mt2-plus";
15566
+ readonly inputPrice: "0.074e-6";
15567
+ readonly outputPrice: "0.295e-6";
15568
+ readonly requestPrice: "0";
15569
+ readonly contextSize: 8192;
15570
+ readonly maxOutput: 4096;
15571
+ readonly streaming: true;
15572
+ readonly vision: false;
15573
+ readonly tools: false;
15574
+ readonly jsonOutput: false;
15575
+ }];
15576
+ }, {
15577
+ readonly id: "hy4-preview";
15578
+ readonly name: "Hy4 Preview";
15579
+ readonly description: "Tencent's preview release of Hy4, its next-generation Hunyuan reasoning model, with a 1M context window and preserved thinking.";
15580
+ readonly family: "tencent";
15581
+ readonly releasedAt: Date;
15582
+ readonly providers: [{
15583
+ readonly providerId: "tencent";
15584
+ readonly externalId: "hy4-preview";
15585
+ readonly inputPrice: "0.834e-6";
15586
+ readonly cachedInputPrice: "0.042e-6";
15587
+ readonly outputPrice: "2.501e-6";
15588
+ readonly requestPrice: "0";
15589
+ readonly contextSize: 1000000;
15590
+ readonly maxOutput: 65536;
15591
+ readonly maxTemperature: 1;
15592
+ readonly streaming: true;
15593
+ readonly reasoning: true;
15594
+ readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
15595
+ readonly vision: false;
15596
+ readonly tools: true;
15597
+ readonly jsonOutput: true;
15598
+ readonly jsonOutputSchema: true;
15255
15599
  }];
15256
15600
  }, {
15257
15601
  readonly id: "nemotron-3-ultra-550b";
@@ -15759,6 +16103,22 @@ export declare const models: [{
15759
16103
  readonly vision: false;
15760
16104
  readonly tools: true;
15761
16105
  readonly jsonOutput: true;
16106
+ }, {
16107
+ readonly providerId: "tencent";
16108
+ readonly externalId: "glm-5.2";
16109
+ readonly inputPrice: "1.4e-6";
16110
+ readonly cachedInputPrice: "0.26e-6";
16111
+ readonly outputPrice: "4.4e-6";
16112
+ readonly requestPrice: "0";
16113
+ readonly contextSize: 1000000;
16114
+ readonly maxOutput: 128000;
16115
+ readonly streaming: true;
16116
+ readonly reasoning: true;
16117
+ readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
16118
+ readonly vision: false;
16119
+ readonly tools: true;
16120
+ readonly jsonOutput: true;
16121
+ readonly jsonOutputSchema: false;
15762
16122
  }];
15763
16123
  }, {
15764
16124
  readonly id: "glm-5.2-fast";
@@ -15899,6 +16259,22 @@ export declare const models: [{
15899
16259
  readonly vision: false;
15900
16260
  readonly tools: true;
15901
16261
  readonly jsonOutput: true;
16262
+ }, {
16263
+ readonly providerId: "tencent";
16264
+ readonly externalId: "glm-5.1";
16265
+ readonly inputPrice: "1.4e-6";
16266
+ readonly cachedInputPrice: "0.26e-6";
16267
+ readonly outputPrice: "4.4e-6";
16268
+ readonly requestPrice: "0";
16269
+ readonly contextSize: 200000;
16270
+ readonly maxOutput: 128000;
16271
+ readonly streaming: true;
16272
+ readonly reasoning: true;
16273
+ readonly reasoningEfforts: ["low", "medium", "high", "xhigh", "max"];
16274
+ readonly vision: false;
16275
+ readonly tools: true;
16276
+ readonly jsonOutput: true;
16277
+ readonly jsonOutputSchema: false;
15902
16278
  }];
15903
16279
  }, {
15904
16280
  readonly id: "glm-5";
@@ -16041,6 +16417,22 @@ export declare const models: [{
16041
16417
  readonly requiresDisableThinkingParam: true;
16042
16418
  readonly vision: false;
16043
16419
  readonly tools: true;
16420
+ }, {
16421
+ readonly providerId: "tencent";
16422
+ readonly externalId: "glm-5";
16423
+ readonly inputPrice: "1e-6";
16424
+ readonly cachedInputPrice: "0.2e-6";
16425
+ readonly outputPrice: "3.2e-6";
16426
+ readonly requestPrice: "0";
16427
+ readonly contextSize: 200000;
16428
+ readonly maxOutput: 128000;
16429
+ readonly streaming: true;
16430
+ readonly reasoning: true;
16431
+ readonly reasoningEfforts: ["none", "minimal", "low", "medium", "high", "xhigh", "max"];
16432
+ readonly vision: false;
16433
+ readonly tools: true;
16434
+ readonly jsonOutput: true;
16435
+ readonly jsonOutputSchema: false;
16044
16436
  }];
16045
16437
  }, {
16046
16438
  readonly id: "glm-4.5";
@@ -16666,6 +17058,52 @@ export declare const models: [{
16666
17058
  readonly jsonOutput: false;
16667
17059
  readonly imageGenerations: true;
16668
17060
  }];
17061
+ }, {
17062
+ readonly id: "glm-5-turbo";
17063
+ readonly name: "GLM-5 Turbo";
17064
+ readonly description: "Zhipu's speed-optimised GLM-5 variant with a 200K context window and deep reasoning.";
17065
+ readonly family: "zai";
17066
+ readonly releasedAt: Date;
17067
+ readonly providers: [{
17068
+ readonly providerId: "tencent";
17069
+ readonly externalId: "glm-5-turbo";
17070
+ readonly inputPrice: "1.2e-6";
17071
+ readonly cachedInputPrice: "0.24e-6";
17072
+ readonly outputPrice: "4e-6";
17073
+ readonly requestPrice: "0";
17074
+ readonly contextSize: 200000;
17075
+ readonly maxOutput: 128000;
17076
+ readonly streaming: true;
17077
+ readonly reasoning: true;
17078
+ readonly reasoningEfforts: ["low", "medium", "high", "xhigh", "max"];
17079
+ readonly vision: false;
17080
+ readonly tools: true;
17081
+ readonly jsonOutput: true;
17082
+ readonly jsonOutputSchema: false;
17083
+ }];
17084
+ }, {
17085
+ readonly id: "glm-5v-turbo";
17086
+ readonly name: "GLM-5V Turbo";
17087
+ readonly description: "Zhipu's speed-optimised multimodal GLM-5 variant that accepts image input alongside text, with a 200K context window.";
17088
+ readonly family: "zai";
17089
+ readonly releasedAt: Date;
17090
+ readonly providers: [{
17091
+ readonly providerId: "tencent";
17092
+ readonly externalId: "glm-5v-turbo";
17093
+ readonly inputPrice: "1.2e-6";
17094
+ readonly cachedInputPrice: "0.24e-6";
17095
+ readonly outputPrice: "4e-6";
17096
+ readonly requestPrice: "0";
17097
+ readonly contextSize: 200000;
17098
+ readonly maxOutput: 128000;
17099
+ readonly streaming: true;
17100
+ readonly reasoning: true;
17101
+ readonly reasoningEfforts: ["low", "medium", "high", "xhigh", "max"];
17102
+ readonly vision: true;
17103
+ readonly tools: true;
17104
+ readonly jsonOutput: true;
17105
+ readonly jsonOutputSchema: false;
17106
+ }];
16669
17107
  }, {
16670
17108
  readonly id: "eleven-multilingual-v2";
16671
17109
  readonly name: "Eleven Multilingual v2";
package/dist/models.js CHANGED
@@ -8,6 +8,7 @@ import { deepseekModels } from "./models/deepseek.js";
8
8
  import { elevenlabsModels } from "./models/elevenlabs.js";
9
9
  import { googleModels } from "./models/google.js";
10
10
  import { inclusionaiModels } from "./models/inclusionai.js";
11
+ import { kinfraModels } from "./models/kinfra.js";
11
12
  import { llmgatewayModels } from "./models/llmgateway.js";
12
13
  import { metaModels } from "./models/meta.js";
13
14
  import { microsoftModels } from "./models/microsoft.js";
@@ -54,6 +55,7 @@ export const models = [
54
55
  ...nousresearchModels,
55
56
  ...reveModels,
56
57
  ...sakanaModels,
58
+ ...kinfraModels,
57
59
  ...tencentModels,
58
60
  ...nvidiaModels,
59
61
  ...openbmbModels,
@@ -1 +1 @@
1
- {"version":3,"file":"models.js","sourceRoot":"","sources":["../src/models.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAC5D,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AAC5C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AAuB5C,MAAM,CAAC,MAAM,oBAAoB,GAAG;IACnC,iBAAiB;IACjB,yBAAyB;IACzB,kBAAkB;IAClB,eAAe;CACN,CAAC;AAq0BX,MAAM,CAAC,MAAM,MAAM,GAAG;IACrB,GAAG,gBAAgB;IACnB,GAAG,YAAY;IACf,GAAG,eAAe;IAClB,GAAG,YAAY;IACf,GAAG,iBAAiB;IACpB,GAAG,gBAAgB;IACnB,GAAG,SAAS;IACZ,GAAG,YAAY;IACf,GAAG,UAAU;IACb,GAAG,cAAc;IACjB,GAAG,aAAa;IAChB,GAAG,eAAe;IAClB,GAAG,aAAa;IAChB,GAAG,cAAc;IACjB,GAAG,aAAa;IAChB,GAAG,gBAAgB;IACnB,GAAG,UAAU;IACb,GAAG,WAAW;IACd,GAAG,eAAe;IAClB,GAAG,kBAAkB;IACrB,GAAG,UAAU;IACb,GAAG,YAAY;IACf,GAAG,aAAa;IAChB,GAAG,YAAY;IACf,GAAG,aAAa;IAChB,GAAG,SAAS;IACZ,GAAG,gBAAgB;CACkB,CAAC"}
1
+ {"version":3,"file":"models.js","sourceRoot":"","sources":["../src/models.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAC5D,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AAC5C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AAuB5C,MAAM,CAAC,MAAM,oBAAoB,GAAG;IACnC,iBAAiB;IACjB,yBAAyB;IACzB,kBAAkB;IAClB,eAAe;CACN,CAAC;AAu0BX,MAAM,CAAC,MAAM,MAAM,GAAG;IACrB,GAAG,gBAAgB;IACnB,GAAG,YAAY;IACf,GAAG,eAAe;IAClB,GAAG,YAAY;IACf,GAAG,iBAAiB;IACpB,GAAG,gBAAgB;IACnB,GAAG,SAAS;IACZ,GAAG,YAAY;IACf,GAAG,UAAU;IACb,GAAG,cAAc;IACjB,GAAG,aAAa;IAChB,GAAG,eAAe;IAClB,GAAG,aAAa;IAChB,GAAG,cAAc;IACjB,GAAG,aAAa;IAChB,GAAG,gBAAgB;IACnB,GAAG,UAAU;IACb,GAAG,WAAW;IACd,GAAG,eAAe;IAClB,GAAG,kBAAkB;IACrB,GAAG,UAAU;IACb,GAAG,YAAY;IACf,GAAG,YAAY;IACf,GAAG,aAAa;IAChB,GAAG,YAAY;IACf,GAAG,aAAa;IAChB,GAAG,SAAS;IACZ,GAAG,gBAAgB;CACkB,CAAC"}
package/dist/providers.js CHANGED
@@ -772,7 +772,7 @@ export const providers = [
772
772
  id: "aws-mantle",
773
773
  name: "AWS Mantle",
774
774
  forwardsSafetyIdentifier: false,
775
- description: "Amazon Bedrock Mantle - OpenAI frontier models served on AWS via the Responses API",
775
+ description: "OpenAI frontier models on Amazon Bedrock's Mantle and Runtime Responses APIs",
776
776
  env: {
777
777
  required: {
778
778
  apiKey: "LLM_AWS_MANTLE_API_KEY",
@@ -793,15 +793,20 @@ export const providers = [
793
793
  optionsKey: "aws_mantle_region",
794
794
  defaultRegion: "us-east-1",
795
795
  regions: [
796
+ { id: "global", label: "Global" },
797
+ { id: "us", label: "US cross-region" },
796
798
  { id: "us-east-1", label: "US East (N. Virginia)" },
797
799
  { id: "us-east-2", label: "US East (Ohio)" },
798
800
  { id: "us-west-2", label: "US West (Oregon)" },
799
801
  ],
800
802
  endpointMap: {
803
+ global: "https://bedrock-runtime.us-east-1.amazonaws.com",
804
+ us: "https://bedrock-runtime.us-east-1.amazonaws.com",
801
805
  "us-east-1": "https://bedrock-mantle.us-east-1.api.aws",
802
806
  "us-east-2": "https://bedrock-mantle.us-east-2.api.aws",
803
807
  "us-west-2": "https://bedrock-mantle.us-west-2.api.aws",
804
808
  },
809
+ modelPrefixMap: { global: "global.", us: "us." },
805
810
  sharedCredentialAcrossRegions: true,
806
811
  },
807
812
  apiKeyInstructions: "Use AWS Bedrock Long-Term API Keys (not IAM service account or private keys)",
@@ -1768,6 +1773,34 @@ export const providers = [
1768
1773
  },
1769
1774
  ],
1770
1775
  },
1776
+ {
1777
+ id: "tencent",
1778
+ name: "Tencent Cloud",
1779
+ forwardsSafetyIdentifier: false,
1780
+ description: "Tencent Cloud's TokenHub model gateway, serving Tencent's own Hunyuan models alongside third-party models through a single OpenAI-compatible API.",
1781
+ env: {
1782
+ required: {
1783
+ apiKey: "LLM_TENCENT_API_KEY",
1784
+ },
1785
+ },
1786
+ streaming: true,
1787
+ cancellation: true,
1788
+ color: "#0052D9",
1789
+ website: "https://www.tencentcloud.com/act/pro/tokenhub",
1790
+ statusPageUrl: null,
1791
+ announcement: null,
1792
+ termsUrl: "https://www.tencentcloud.com/document/product/301/78869",
1793
+ privacyPolicyUrl: "https://www.tencentcloud.com/document/product/301/17345",
1794
+ usagePolicyUrl: "https://www.tencentcloud.com/document/product/301/9245",
1795
+ legalEntity: "Tencent Cloud LLC.",
1796
+ headquarters: "CN",
1797
+ dataPolicy: {
1798
+ apiTraining: false,
1799
+ promptLogging: false,
1800
+ retentionPeriod: null,
1801
+ },
1802
+ priority: 1.2,
1803
+ },
1771
1804
  ];
1772
1805
  export function getProviderDefinition(providerId) {
1773
1806
  return providers.find((p) => p.id === providerId);