@flame0510/project-aether 1.6.0 → 1.6.2
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/app/api/update-check/route.ts +4 -1
- package/app/components/VersionBanner.tsx +80 -52
- package/docs/dev/API-REFERENCE.md +4 -0
- package/docs/dev/GATEWAY.md +23 -0
- package/models.config.json +937 -45
- package/package.json +4 -2
- package/scripts/model-info-verify.mjs +194 -0
package/models.config.json
CHANGED
|
@@ -145,7 +145,12 @@
|
|
|
145
145
|
"name": "DeepSeek V4 Pro",
|
|
146
146
|
"provider": "deepseek",
|
|
147
147
|
"enabled": true,
|
|
148
|
-
"modality": "text->text"
|
|
148
|
+
"modality": "text->text",
|
|
149
|
+
"info": {
|
|
150
|
+
"params": "DeepSeek-V4-Pro with 1.6T parameters (49B activated)",
|
|
151
|
+
"docUrl": "https://arxiv.org/abs/2606.19348",
|
|
152
|
+
"notes": "DeepSeek-V4-Pro uses a hybrid attention mechanism combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA)."
|
|
153
|
+
}
|
|
149
154
|
},
|
|
150
155
|
{
|
|
151
156
|
"id": "groq/llama-3.1-8b-instant",
|
|
@@ -177,7 +182,32 @@
|
|
|
177
182
|
"name": "Kimi K2.7 Code",
|
|
178
183
|
"provider": "kimi",
|
|
179
184
|
"enabled": false,
|
|
180
|
-
"modality": "text+image->text"
|
|
185
|
+
"modality": "text+image->text",
|
|
186
|
+
"info": {
|
|
187
|
+
"params": "1T total parameters, 32B activated parameters",
|
|
188
|
+
"benchmarks": [
|
|
189
|
+
{
|
|
190
|
+
"name": "Kimi Code Bench v2",
|
|
191
|
+
"value": 62.0,
|
|
192
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.7-Code",
|
|
193
|
+
"asOf": "2026-06-11"
|
|
194
|
+
},
|
|
195
|
+
{
|
|
196
|
+
"name": "MCP Atlas",
|
|
197
|
+
"value": 76.0,
|
|
198
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.7-Code",
|
|
199
|
+
"asOf": "2026-06-11"
|
|
200
|
+
},
|
|
201
|
+
{
|
|
202
|
+
"name": "Program Bench",
|
|
203
|
+
"value": 53.6,
|
|
204
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.7-Code",
|
|
205
|
+
"asOf": "2026-06-11"
|
|
206
|
+
}
|
|
207
|
+
],
|
|
208
|
+
"docUrl": "https://platform.moonshot.ai",
|
|
209
|
+
"notes": "Kimi K2.7 Code includes a MoonViT vision encoder (400M parameters) and supports image and video input."
|
|
210
|
+
}
|
|
181
211
|
},
|
|
182
212
|
{
|
|
183
213
|
"id": "kimi/kimi-k2.7-code-highspeed",
|
|
@@ -191,7 +221,32 @@
|
|
|
191
221
|
"name": "Kimi K2.6",
|
|
192
222
|
"provider": "kimi",
|
|
193
223
|
"enabled": false,
|
|
194
|
-
"modality": "text+image->text"
|
|
224
|
+
"modality": "text+image->text",
|
|
225
|
+
"info": {
|
|
226
|
+
"params": "1T total parameters, 32B activated parameters",
|
|
227
|
+
"benchmarks": [
|
|
228
|
+
{
|
|
229
|
+
"name": "SWE-Bench Verified",
|
|
230
|
+
"value": 80.2,
|
|
231
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.6",
|
|
232
|
+
"asOf": "2026-04-14"
|
|
233
|
+
},
|
|
234
|
+
{
|
|
235
|
+
"name": "GPQA-Diamond",
|
|
236
|
+
"value": 90.5,
|
|
237
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.6",
|
|
238
|
+
"asOf": "2026-04-14"
|
|
239
|
+
},
|
|
240
|
+
{
|
|
241
|
+
"name": "AIME 2026",
|
|
242
|
+
"value": 96.4,
|
|
243
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.6",
|
|
244
|
+
"asOf": "2026-04-14"
|
|
245
|
+
}
|
|
246
|
+
],
|
|
247
|
+
"docUrl": "https://platform.moonshot.ai",
|
|
248
|
+
"notes": "Kimi K2.6 is a native multimodal agentic model with a MoonViT vision encoder (400M parameters)."
|
|
249
|
+
}
|
|
195
250
|
},
|
|
196
251
|
{
|
|
197
252
|
"id": "glm/glm-5.3",
|
|
@@ -242,35 +297,156 @@
|
|
|
242
297
|
"name": "GLM-5.2",
|
|
243
298
|
"provider": "glm",
|
|
244
299
|
"enabled": false,
|
|
245
|
-
"modality": "text->text"
|
|
300
|
+
"modality": "text->text",
|
|
301
|
+
"info": {
|
|
302
|
+
"benchmarks": [
|
|
303
|
+
{
|
|
304
|
+
"name": "AIME 2026",
|
|
305
|
+
"value": 99.2,
|
|
306
|
+
"source": "https://huggingface.co/zai-org/GLM-5.2",
|
|
307
|
+
"asOf": "2026-06-16"
|
|
308
|
+
},
|
|
309
|
+
{
|
|
310
|
+
"name": "GPQA-Diamond",
|
|
311
|
+
"value": 91.2,
|
|
312
|
+
"source": "https://huggingface.co/zai-org/GLM-5.2",
|
|
313
|
+
"asOf": "2026-06-16"
|
|
314
|
+
},
|
|
315
|
+
{
|
|
316
|
+
"name": "SWE-bench Pro",
|
|
317
|
+
"value": 62.1,
|
|
318
|
+
"source": "https://huggingface.co/zai-org/GLM-5.2",
|
|
319
|
+
"asOf": "2026-06-16"
|
|
320
|
+
}
|
|
321
|
+
],
|
|
322
|
+
"docUrl": "https://docs.z.ai/guides/llm/glm-5.2",
|
|
323
|
+
"notes": "GLM-5.2 delivers a solid 1M-token context and proposes IndexShare, which reuses the same indexer across every four sparse attention layers, reducing per-token FLOPs by 2.9× at 1M context."
|
|
324
|
+
}
|
|
246
325
|
},
|
|
247
326
|
{
|
|
248
327
|
"id": "glm/glm-5.1",
|
|
249
328
|
"name": "GLM-5.1",
|
|
250
329
|
"provider": "glm",
|
|
251
330
|
"enabled": false,
|
|
252
|
-
"modality": "text->text"
|
|
331
|
+
"modality": "text->text",
|
|
332
|
+
"info": {
|
|
333
|
+
"benchmarks": [
|
|
334
|
+
{
|
|
335
|
+
"name": "GPQA-Diamond",
|
|
336
|
+
"value": 86.2,
|
|
337
|
+
"source": "https://huggingface.co/zai-org/GLM-5.1",
|
|
338
|
+
"asOf": "2026-04-03"
|
|
339
|
+
},
|
|
340
|
+
{
|
|
341
|
+
"name": "SWE-Bench Pro",
|
|
342
|
+
"value": 58.4,
|
|
343
|
+
"source": "https://huggingface.co/zai-org/GLM-5.1",
|
|
344
|
+
"asOf": "2026-04-03"
|
|
345
|
+
},
|
|
346
|
+
{
|
|
347
|
+
"name": "HLE",
|
|
348
|
+
"value": 31.0,
|
|
349
|
+
"source": "https://huggingface.co/zai-org/GLM-5.1",
|
|
350
|
+
"asOf": "2026-04-03"
|
|
351
|
+
}
|
|
352
|
+
],
|
|
353
|
+
"docUrl": "https://docs.z.ai/guides/llm/glm-5.1",
|
|
354
|
+
"notes": "GLM-5.1 sustains optimization over hundreds of rounds and thousands of tool calls, staying effective on agentic tasks over long horizons."
|
|
355
|
+
}
|
|
253
356
|
},
|
|
254
357
|
{
|
|
255
358
|
"id": "glm/glm-5",
|
|
256
359
|
"name": "GLM-5",
|
|
257
360
|
"provider": "glm",
|
|
258
361
|
"enabled": false,
|
|
259
|
-
"modality": "text->text"
|
|
362
|
+
"modality": "text->text",
|
|
363
|
+
"info": {
|
|
364
|
+
"params": "GLM-5 scales from 355B parameters (32B active) to 744B parameters (40B active)",
|
|
365
|
+
"benchmarks": [
|
|
366
|
+
{
|
|
367
|
+
"name": "GPQA-Diamond",
|
|
368
|
+
"value": 86.0,
|
|
369
|
+
"source": "https://huggingface.co/zai-org/GLM-5",
|
|
370
|
+
"asOf": "2026-02-11"
|
|
371
|
+
},
|
|
372
|
+
{
|
|
373
|
+
"name": "SWE-bench Verified",
|
|
374
|
+
"value": 77.8,
|
|
375
|
+
"source": "https://huggingface.co/zai-org/GLM-5",
|
|
376
|
+
"asOf": "2026-02-11"
|
|
377
|
+
},
|
|
378
|
+
{
|
|
379
|
+
"name": "HLE",
|
|
380
|
+
"value": 30.5,
|
|
381
|
+
"source": "https://huggingface.co/zai-org/GLM-5",
|
|
382
|
+
"asOf": "2026-02-11"
|
|
383
|
+
}
|
|
384
|
+
],
|
|
385
|
+
"docUrl": "https://docs.z.ai/guides/llm/glm-5",
|
|
386
|
+
"notes": "GLM-5 integrates DeepSeek Sparse Attention (DSA), largely reducing deployment cost while preserving long-context capacity."
|
|
387
|
+
}
|
|
260
388
|
},
|
|
261
389
|
{
|
|
262
390
|
"id": "glm/glm-4.7",
|
|
263
391
|
"name": "GLM-4.7",
|
|
264
392
|
"provider": "glm",
|
|
265
393
|
"enabled": false,
|
|
266
|
-
"modality": "text->text"
|
|
394
|
+
"modality": "text->text",
|
|
395
|
+
"info": {
|
|
396
|
+
"benchmarks": [
|
|
397
|
+
{
|
|
398
|
+
"name": "MMLU-Pro",
|
|
399
|
+
"value": 84.3,
|
|
400
|
+
"source": "https://huggingface.co/zai-org/GLM-4.7",
|
|
401
|
+
"asOf": "2025-12-22"
|
|
402
|
+
},
|
|
403
|
+
{
|
|
404
|
+
"name": "GPQA-Diamond",
|
|
405
|
+
"value": 85.7,
|
|
406
|
+
"source": "https://huggingface.co/zai-org/GLM-4.7",
|
|
407
|
+
"asOf": "2025-12-22"
|
|
408
|
+
},
|
|
409
|
+
{
|
|
410
|
+
"name": "AIME 2025",
|
|
411
|
+
"value": 95.7,
|
|
412
|
+
"source": "https://huggingface.co/zai-org/GLM-4.7",
|
|
413
|
+
"asOf": "2025-12-22"
|
|
414
|
+
}
|
|
415
|
+
],
|
|
416
|
+
"docUrl": "https://docs.z.ai/guides/llm/glm-4.7",
|
|
417
|
+
"notes": "GLM-4.7 adds Preserved Thinking, retaining thinking blocks across multi-turn coding conversations."
|
|
418
|
+
}
|
|
267
419
|
},
|
|
268
420
|
{
|
|
269
421
|
"id": "glm/glm-4.7-flash",
|
|
270
422
|
"name": "GLM-4.7 Flash",
|
|
271
423
|
"provider": "glm",
|
|
272
424
|
"enabled": false,
|
|
273
|
-
"modality": "text->text"
|
|
425
|
+
"modality": "text->text",
|
|
426
|
+
"info": {
|
|
427
|
+
"params": "GLM-4.7-Flash is a 30B-A3B MoE model.",
|
|
428
|
+
"benchmarks": [
|
|
429
|
+
{
|
|
430
|
+
"name": "AIME 25",
|
|
431
|
+
"value": 91.6,
|
|
432
|
+
"source": "https://huggingface.co/zai-org/GLM-4.7-Flash",
|
|
433
|
+
"asOf": "2026-01-19"
|
|
434
|
+
},
|
|
435
|
+
{
|
|
436
|
+
"name": "GPQA",
|
|
437
|
+
"value": 75.2,
|
|
438
|
+
"source": "https://huggingface.co/zai-org/GLM-4.7-Flash",
|
|
439
|
+
"asOf": "2026-01-19"
|
|
440
|
+
},
|
|
441
|
+
{
|
|
442
|
+
"name": "SWE-bench Verified",
|
|
443
|
+
"value": 59.2,
|
|
444
|
+
"source": "https://huggingface.co/zai-org/GLM-4.7-Flash",
|
|
445
|
+
"asOf": "2026-01-19"
|
|
446
|
+
}
|
|
447
|
+
],
|
|
448
|
+
"docUrl": "https://docs.z.ai/guides/llm/glm-4.7"
|
|
449
|
+
}
|
|
274
450
|
},
|
|
275
451
|
{
|
|
276
452
|
"id": "glm/glm-4.7-flashx",
|
|
@@ -340,14 +516,24 @@
|
|
|
340
516
|
"name": "DeepSeek V4 Flash (OR)",
|
|
341
517
|
"provider": "openrouter",
|
|
342
518
|
"enabled": false,
|
|
343
|
-
"modality": "text->text"
|
|
519
|
+
"modality": "text->text",
|
|
520
|
+
"info": {
|
|
521
|
+
"params": "DeepSeek-V4-Flash with 284B parameters (13B activated)",
|
|
522
|
+
"docUrl": "https://arxiv.org/abs/2606.19348",
|
|
523
|
+
"notes": "DeepSeek-V4-Flash is a Mixture-of-Experts model supporting a one-million-token context length."
|
|
524
|
+
}
|
|
344
525
|
},
|
|
345
526
|
{
|
|
346
527
|
"id": "openrouter/deepseek/deepseek-v4-pro",
|
|
347
528
|
"name": "DeepSeek V4 Pro (OR)",
|
|
348
529
|
"provider": "openrouter",
|
|
349
530
|
"enabled": false,
|
|
350
|
-
"modality": "text->text"
|
|
531
|
+
"modality": "text->text",
|
|
532
|
+
"info": {
|
|
533
|
+
"params": "DeepSeek-V4-Pro with 1.6T parameters (49B activated)",
|
|
534
|
+
"docUrl": "https://arxiv.org/abs/2606.19348",
|
|
535
|
+
"notes": "DeepSeek-V4-Pro uses a hybrid attention mechanism combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA)."
|
|
536
|
+
}
|
|
351
537
|
},
|
|
352
538
|
{
|
|
353
539
|
"id": "openrouter/google/gemini-2.5-flash",
|
|
@@ -417,21 +603,76 @@
|
|
|
417
603
|
"name": "Qwen 3.6 27B (OR)",
|
|
418
604
|
"provider": "openrouter",
|
|
419
605
|
"enabled": true,
|
|
420
|
-
"modality": "text+image+video->text"
|
|
606
|
+
"modality": "text+image+video->text",
|
|
607
|
+
"info": {
|
|
608
|
+
"params": "27B parameters",
|
|
609
|
+
"benchmarks": [
|
|
610
|
+
{
|
|
611
|
+
"name": "SWE-bench Verified",
|
|
612
|
+
"value": 77.2,
|
|
613
|
+
"source": "https://huggingface.co/Qwen/Qwen3.6-27B",
|
|
614
|
+
"asOf": "2026-04-21"
|
|
615
|
+
},
|
|
616
|
+
{
|
|
617
|
+
"name": "GPQA Diamond",
|
|
618
|
+
"value": 87.8,
|
|
619
|
+
"source": "https://huggingface.co/Qwen/Qwen3.6-27B",
|
|
620
|
+
"asOf": "2026-04-21"
|
|
621
|
+
},
|
|
622
|
+
{
|
|
623
|
+
"name": "MMLU-Pro",
|
|
624
|
+
"value": 86.2,
|
|
625
|
+
"source": "https://huggingface.co/Qwen/Qwen3.6-27B",
|
|
626
|
+
"asOf": "2026-04-21"
|
|
627
|
+
}
|
|
628
|
+
],
|
|
629
|
+
"docUrl": "https://github.com/QwenLM/Qwen3.8",
|
|
630
|
+
"notes": "Qwen3.6-27B is a Causal Language Model with a vision encoder that introduces thinking preservation to retain reasoning context from historical messages."
|
|
631
|
+
}
|
|
421
632
|
},
|
|
422
633
|
{
|
|
423
634
|
"id": "openrouter/qwen/qwen3.6-35b-a3b",
|
|
424
635
|
"name": "Qwen 3.6 35B-A3B (OR)",
|
|
425
636
|
"provider": "openrouter",
|
|
426
637
|
"enabled": true,
|
|
427
|
-
"modality": "text+image+video->text"
|
|
638
|
+
"modality": "text+image+video->text",
|
|
639
|
+
"info": {
|
|
640
|
+
"params": "Number of Parameters: 35B in total and 3B activated",
|
|
641
|
+
"benchmarks": [
|
|
642
|
+
{
|
|
643
|
+
"name": "SWE-bench Verified",
|
|
644
|
+
"value": 73.4,
|
|
645
|
+
"source": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B",
|
|
646
|
+
"asOf": "2026-04-15"
|
|
647
|
+
},
|
|
648
|
+
{
|
|
649
|
+
"name": "MMLU-Pro",
|
|
650
|
+
"value": 85.2,
|
|
651
|
+
"source": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B",
|
|
652
|
+
"asOf": "2026-04-15"
|
|
653
|
+
},
|
|
654
|
+
{
|
|
655
|
+
"name": "GPQA",
|
|
656
|
+
"value": 86.0,
|
|
657
|
+
"source": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B",
|
|
658
|
+
"asOf": "2026-04-15"
|
|
659
|
+
}
|
|
660
|
+
],
|
|
661
|
+
"docUrl": "https://qwen.ai/blog?id=qwen3.6-35b-a3b",
|
|
662
|
+
"notes": "Qwen3.6-35B-A3B is a causal language model with a vision encoder, i.e. multimodal."
|
|
663
|
+
}
|
|
428
664
|
},
|
|
429
665
|
{
|
|
430
666
|
"id": "openrouter/qwen/qwen3-coder",
|
|
431
667
|
"name": "Qwen 3 Coder (OR)",
|
|
432
668
|
"provider": "openrouter",
|
|
433
669
|
"enabled": true,
|
|
434
|
-
"modality": "text->text"
|
|
670
|
+
"modality": "text->text",
|
|
671
|
+
"info": {
|
|
672
|
+
"params": "480B in total and 35B activated",
|
|
673
|
+
"docUrl": "https://github.com/QwenLM/Qwen3-Coder",
|
|
674
|
+
"notes": "Qwen3-Coder-480B-A35B-Instruct supports only non-thinking mode and does not generate <think></think> blocks."
|
|
675
|
+
}
|
|
435
676
|
},
|
|
436
677
|
{
|
|
437
678
|
"id": "openrouter/qwen/qwen3-max",
|
|
@@ -452,49 +693,201 @@
|
|
|
452
693
|
"name": "Qwen 3 Next 80B Instruct (OR)",
|
|
453
694
|
"provider": "openrouter",
|
|
454
695
|
"enabled": true,
|
|
455
|
-
"modality": "text->text"
|
|
696
|
+
"modality": "text->text",
|
|
697
|
+
"info": {
|
|
698
|
+
"params": "Number of Parameters: 80B in total and 3B activated",
|
|
699
|
+
"benchmarks": [
|
|
700
|
+
{
|
|
701
|
+
"name": "MMLU-Pro",
|
|
702
|
+
"value": 80.6,
|
|
703
|
+
"source": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct",
|
|
704
|
+
"asOf": "2025-09-09"
|
|
705
|
+
},
|
|
706
|
+
{
|
|
707
|
+
"name": "GPQA",
|
|
708
|
+
"value": 72.9,
|
|
709
|
+
"source": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct",
|
|
710
|
+
"asOf": "2025-09-09"
|
|
711
|
+
},
|
|
712
|
+
{
|
|
713
|
+
"name": "AIME25",
|
|
714
|
+
"value": 69.5,
|
|
715
|
+
"source": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Instruct",
|
|
716
|
+
"asOf": "2025-09-09"
|
|
717
|
+
}
|
|
718
|
+
],
|
|
719
|
+
"notes": "Qwen3-Next-80B-A3B-Instruct supports only instruct (non-thinking) mode."
|
|
720
|
+
}
|
|
456
721
|
},
|
|
457
722
|
{
|
|
458
723
|
"id": "openrouter/qwen/qwen3-next-80b-a3b-thinking",
|
|
459
724
|
"name": "Qwen 3 Next 80B Thinking (OR)",
|
|
460
725
|
"provider": "openrouter",
|
|
461
726
|
"enabled": true,
|
|
462
|
-
"modality": "text->text"
|
|
727
|
+
"modality": "text->text",
|
|
728
|
+
"info": {
|
|
729
|
+
"params": "Number of Parameters: 80B in total and 3B activated",
|
|
730
|
+
"benchmarks": [
|
|
731
|
+
{
|
|
732
|
+
"name": "MMLU-Pro",
|
|
733
|
+
"value": 82.7,
|
|
734
|
+
"source": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking",
|
|
735
|
+
"asOf": "2025-09-09"
|
|
736
|
+
},
|
|
737
|
+
{
|
|
738
|
+
"name": "GPQA",
|
|
739
|
+
"value": 77.2,
|
|
740
|
+
"source": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking",
|
|
741
|
+
"asOf": "2025-09-09"
|
|
742
|
+
},
|
|
743
|
+
{
|
|
744
|
+
"name": "AIME25",
|
|
745
|
+
"value": 87.8,
|
|
746
|
+
"source": "https://huggingface.co/Qwen/Qwen3-Next-80B-A3B-Thinking",
|
|
747
|
+
"asOf": "2025-09-09"
|
|
748
|
+
}
|
|
749
|
+
],
|
|
750
|
+
"notes": "Qwen3-Next-80B-A3B-Thinking supports only thinking mode."
|
|
751
|
+
}
|
|
463
752
|
},
|
|
464
753
|
{
|
|
465
754
|
"id": "openrouter/minimax/minimax-m3",
|
|
466
755
|
"name": "Minimax M3 (OR)",
|
|
467
756
|
"provider": "openrouter",
|
|
468
757
|
"enabled": true,
|
|
469
|
-
"modality": "text+image+video->text"
|
|
758
|
+
"modality": "text+image+video->text",
|
|
759
|
+
"info": {
|
|
760
|
+
"params": "It has ~428B parameters and ~23B activated parameters.",
|
|
761
|
+
"docUrl": "https://platform.minimax.io/docs/guides/text-generation",
|
|
762
|
+
"notes": "MiniMax-M3 uses MiniMax Sparse Attention (MSA) for efficient million-token context processing."
|
|
763
|
+
}
|
|
470
764
|
},
|
|
471
765
|
{
|
|
472
766
|
"id": "openrouter/moonshotai/kimi-k2.7-code",
|
|
473
767
|
"name": "Kimi K2.7 Code (OR)",
|
|
474
768
|
"provider": "openrouter",
|
|
475
769
|
"enabled": true,
|
|
476
|
-
"modality": "text+image->text"
|
|
770
|
+
"modality": "text+image->text",
|
|
771
|
+
"info": {
|
|
772
|
+
"params": "1T total parameters, 32B activated parameters",
|
|
773
|
+
"benchmarks": [
|
|
774
|
+
{
|
|
775
|
+
"name": "Kimi Code Bench v2",
|
|
776
|
+
"value": 62.0,
|
|
777
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.7-Code",
|
|
778
|
+
"asOf": "2026-06-11"
|
|
779
|
+
},
|
|
780
|
+
{
|
|
781
|
+
"name": "MCP Atlas",
|
|
782
|
+
"value": 76.0,
|
|
783
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.7-Code",
|
|
784
|
+
"asOf": "2026-06-11"
|
|
785
|
+
},
|
|
786
|
+
{
|
|
787
|
+
"name": "Program Bench",
|
|
788
|
+
"value": 53.6,
|
|
789
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.7-Code",
|
|
790
|
+
"asOf": "2026-06-11"
|
|
791
|
+
}
|
|
792
|
+
],
|
|
793
|
+
"docUrl": "https://platform.moonshot.ai",
|
|
794
|
+
"notes": "Kimi K2.7 Code includes a MoonViT vision encoder (400M parameters) and supports image and video input."
|
|
795
|
+
}
|
|
477
796
|
},
|
|
478
797
|
{
|
|
479
798
|
"id": "openrouter/moonshotai/kimi-k2.6",
|
|
480
799
|
"name": "Kimi K2.6 (OR)",
|
|
481
800
|
"provider": "openrouter",
|
|
482
801
|
"enabled": true,
|
|
483
|
-
"modality": "text+image->text"
|
|
802
|
+
"modality": "text+image->text",
|
|
803
|
+
"info": {
|
|
804
|
+
"params": "1T total parameters, 32B activated parameters",
|
|
805
|
+
"benchmarks": [
|
|
806
|
+
{
|
|
807
|
+
"name": "SWE-Bench Verified",
|
|
808
|
+
"value": 80.2,
|
|
809
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.6",
|
|
810
|
+
"asOf": "2026-04-14"
|
|
811
|
+
},
|
|
812
|
+
{
|
|
813
|
+
"name": "GPQA-Diamond",
|
|
814
|
+
"value": 90.5,
|
|
815
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.6",
|
|
816
|
+
"asOf": "2026-04-14"
|
|
817
|
+
},
|
|
818
|
+
{
|
|
819
|
+
"name": "AIME 2026",
|
|
820
|
+
"value": 96.4,
|
|
821
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.6",
|
|
822
|
+
"asOf": "2026-04-14"
|
|
823
|
+
}
|
|
824
|
+
],
|
|
825
|
+
"docUrl": "https://platform.moonshot.ai",
|
|
826
|
+
"notes": "Kimi K2.6 is a native multimodal agentic model with a MoonViT vision encoder (400M parameters)."
|
|
827
|
+
}
|
|
484
828
|
},
|
|
485
829
|
{
|
|
486
830
|
"id": "openrouter/moonshotai/kimi-k2",
|
|
487
831
|
"name": "Kimi K2 (OR)",
|
|
488
832
|
"provider": "openrouter",
|
|
489
833
|
"enabled": true,
|
|
490
|
-
"modality": "text->text"
|
|
834
|
+
"modality": "text->text",
|
|
835
|
+
"info": {
|
|
836
|
+
"params": "Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total parameters.",
|
|
837
|
+
"benchmarks": [
|
|
838
|
+
{
|
|
839
|
+
"name": "LiveCodeBench v6",
|
|
840
|
+
"value": 53.7,
|
|
841
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2-Instruct",
|
|
842
|
+
"asOf": "2025-07-11"
|
|
843
|
+
},
|
|
844
|
+
{
|
|
845
|
+
"name": "MultiPL-E",
|
|
846
|
+
"value": 85.7,
|
|
847
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2-Instruct",
|
|
848
|
+
"asOf": "2025-07-11"
|
|
849
|
+
},
|
|
850
|
+
{
|
|
851
|
+
"name": "SWE-bench Verified (Agentless Coding)",
|
|
852
|
+
"value": 51.8,
|
|
853
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2-Instruct",
|
|
854
|
+
"asOf": "2025-07-11"
|
|
855
|
+
}
|
|
856
|
+
],
|
|
857
|
+
"docUrl": "https://github.com/moonshotai/Kimi-K2",
|
|
858
|
+
"notes": "Kimi-K2-Instruct is a reflex-grade model without long thinking."
|
|
859
|
+
}
|
|
491
860
|
},
|
|
492
861
|
{
|
|
493
862
|
"id": "openrouter/moonshotai/kimi-k2-thinking",
|
|
494
863
|
"name": "Kimi K2 Thinking (OR)",
|
|
495
864
|
"provider": "openrouter",
|
|
496
865
|
"enabled": true,
|
|
497
|
-
"modality": "text->text"
|
|
866
|
+
"modality": "text->text",
|
|
867
|
+
"info": {
|
|
868
|
+
"benchmarks": [
|
|
869
|
+
{
|
|
870
|
+
"name": "AIME25",
|
|
871
|
+
"value": 94.5,
|
|
872
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2-Thinking",
|
|
873
|
+
"asOf": "2025-11-04"
|
|
874
|
+
},
|
|
875
|
+
{
|
|
876
|
+
"name": "MMLU-Pro",
|
|
877
|
+
"value": 84.6,
|
|
878
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2-Thinking",
|
|
879
|
+
"asOf": "2025-11-04"
|
|
880
|
+
},
|
|
881
|
+
{
|
|
882
|
+
"name": "SWE-bench Verified",
|
|
883
|
+
"value": 71.3,
|
|
884
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2-Thinking",
|
|
885
|
+
"asOf": "2025-11-04"
|
|
886
|
+
}
|
|
887
|
+
],
|
|
888
|
+
"docUrl": "https://moonshotai.github.io/Kimi-K2/thinking.html",
|
|
889
|
+
"notes": "Kimi K2 Thinking is a native INT4 quantized model with a 256k context window."
|
|
890
|
+
}
|
|
498
891
|
},
|
|
499
892
|
{
|
|
500
893
|
"id": "openrouter/z-ai/glm-5.3",
|
|
@@ -545,21 +938,94 @@
|
|
|
545
938
|
"name": "GLM 5.2 (OR)",
|
|
546
939
|
"provider": "openrouter",
|
|
547
940
|
"enabled": true,
|
|
548
|
-
"modality": "text->text"
|
|
941
|
+
"modality": "text->text",
|
|
942
|
+
"info": {
|
|
943
|
+
"benchmarks": [
|
|
944
|
+
{
|
|
945
|
+
"name": "AIME 2026",
|
|
946
|
+
"value": 99.2,
|
|
947
|
+
"source": "https://huggingface.co/zai-org/GLM-5.2",
|
|
948
|
+
"asOf": "2026-06-16"
|
|
949
|
+
},
|
|
950
|
+
{
|
|
951
|
+
"name": "GPQA-Diamond",
|
|
952
|
+
"value": 91.2,
|
|
953
|
+
"source": "https://huggingface.co/zai-org/GLM-5.2",
|
|
954
|
+
"asOf": "2026-06-16"
|
|
955
|
+
},
|
|
956
|
+
{
|
|
957
|
+
"name": "SWE-bench Pro",
|
|
958
|
+
"value": 62.1,
|
|
959
|
+
"source": "https://huggingface.co/zai-org/GLM-5.2",
|
|
960
|
+
"asOf": "2026-06-16"
|
|
961
|
+
}
|
|
962
|
+
],
|
|
963
|
+
"docUrl": "https://docs.z.ai/guides/llm/glm-5.2",
|
|
964
|
+
"notes": "GLM-5.2 delivers a solid 1M-token context and proposes IndexShare, which reuses the same indexer across every four sparse attention layers, reducing per-token FLOPs by 2.9× at 1M context."
|
|
965
|
+
}
|
|
549
966
|
},
|
|
550
967
|
{
|
|
551
968
|
"id": "openrouter/z-ai/glm-5.1",
|
|
552
969
|
"name": "GLM 5.1 (OR)",
|
|
553
970
|
"provider": "openrouter",
|
|
554
971
|
"enabled": true,
|
|
555
|
-
"modality": "text->text"
|
|
972
|
+
"modality": "text->text",
|
|
973
|
+
"info": {
|
|
974
|
+
"benchmarks": [
|
|
975
|
+
{
|
|
976
|
+
"name": "GPQA-Diamond",
|
|
977
|
+
"value": 86.2,
|
|
978
|
+
"source": "https://huggingface.co/zai-org/GLM-5.1",
|
|
979
|
+
"asOf": "2026-04-03"
|
|
980
|
+
},
|
|
981
|
+
{
|
|
982
|
+
"name": "SWE-Bench Pro",
|
|
983
|
+
"value": 58.4,
|
|
984
|
+
"source": "https://huggingface.co/zai-org/GLM-5.1",
|
|
985
|
+
"asOf": "2026-04-03"
|
|
986
|
+
},
|
|
987
|
+
{
|
|
988
|
+
"name": "HLE",
|
|
989
|
+
"value": 31.0,
|
|
990
|
+
"source": "https://huggingface.co/zai-org/GLM-5.1",
|
|
991
|
+
"asOf": "2026-04-03"
|
|
992
|
+
}
|
|
993
|
+
],
|
|
994
|
+
"docUrl": "https://docs.z.ai/guides/llm/glm-5.1",
|
|
995
|
+
"notes": "GLM-5.1 sustains optimization over hundreds of rounds and thousands of tool calls, staying effective on agentic tasks over long horizons."
|
|
996
|
+
}
|
|
556
997
|
},
|
|
557
998
|
{
|
|
558
999
|
"id": "openrouter/z-ai/glm-5",
|
|
559
1000
|
"name": "GLM 5 (OR)",
|
|
560
1001
|
"provider": "openrouter",
|
|
561
1002
|
"enabled": true,
|
|
562
|
-
"modality": "text->text"
|
|
1003
|
+
"modality": "text->text",
|
|
1004
|
+
"info": {
|
|
1005
|
+
"params": "GLM-5 scales from 355B parameters (32B active) to 744B parameters (40B active)",
|
|
1006
|
+
"benchmarks": [
|
|
1007
|
+
{
|
|
1008
|
+
"name": "GPQA-Diamond",
|
|
1009
|
+
"value": 86.0,
|
|
1010
|
+
"source": "https://huggingface.co/zai-org/GLM-5",
|
|
1011
|
+
"asOf": "2026-02-11"
|
|
1012
|
+
},
|
|
1013
|
+
{
|
|
1014
|
+
"name": "SWE-bench Verified",
|
|
1015
|
+
"value": 77.8,
|
|
1016
|
+
"source": "https://huggingface.co/zai-org/GLM-5",
|
|
1017
|
+
"asOf": "2026-02-11"
|
|
1018
|
+
},
|
|
1019
|
+
{
|
|
1020
|
+
"name": "HLE",
|
|
1021
|
+
"value": 30.5,
|
|
1022
|
+
"source": "https://huggingface.co/zai-org/GLM-5",
|
|
1023
|
+
"asOf": "2026-02-11"
|
|
1024
|
+
}
|
|
1025
|
+
],
|
|
1026
|
+
"docUrl": "https://docs.z.ai/guides/llm/glm-5",
|
|
1027
|
+
"notes": "GLM-5 integrates DeepSeek Sparse Attention (DSA), largely reducing deployment cost while preserving long-context capacity."
|
|
1028
|
+
}
|
|
563
1029
|
},
|
|
564
1030
|
{
|
|
565
1031
|
"id": "openrouter/z-ai/glm-5v-turbo",
|
|
@@ -594,49 +1060,124 @@
|
|
|
594
1060
|
"name": "StepFun 3.7 Flash (OR)",
|
|
595
1061
|
"provider": "openrouter",
|
|
596
1062
|
"enabled": true,
|
|
597
|
-
"modality": "text+image+video->text"
|
|
1063
|
+
"modality": "text+image+video->text",
|
|
1064
|
+
"info": {
|
|
1065
|
+
"params": "Step 3.7 Flash is a 198B-parameter sparse Mixture-of-Experts (MoE) vision-language model that combines a 196B-parameter language backbone with a 1.8B-parameter vision encoder for native image understanding.",
|
|
1066
|
+
"docUrl": "https://platform.stepfun.ai",
|
|
1067
|
+
"notes": "Step 3.7 Flash is a vision-language model with a 1.8B-parameter vision encoder for native image understanding."
|
|
1068
|
+
}
|
|
598
1069
|
},
|
|
599
1070
|
{
|
|
600
1071
|
"id": "openrouter/stepfun/step-3.5-flash",
|
|
601
1072
|
"name": "StepFun 3.5 Flash (OR)",
|
|
602
1073
|
"provider": "openrouter",
|
|
603
1074
|
"enabled": true,
|
|
604
|
-
"modality": "text->text"
|
|
1075
|
+
"modality": "text->text",
|
|
1076
|
+
"info": {
|
|
1077
|
+
"params": "Built on a sparse Mixture of Experts (MoE) architecture, it selectively activates only 11B of its 196B parameters per token.",
|
|
1078
|
+
"benchmarks": [
|
|
1079
|
+
{
|
|
1080
|
+
"name": "τ²-Bench",
|
|
1081
|
+
"value": 88.2,
|
|
1082
|
+
"source": "https://huggingface.co/stepfun-ai/Step-3.5-Flash",
|
|
1083
|
+
"asOf": "2026-02-01"
|
|
1084
|
+
},
|
|
1085
|
+
{
|
|
1086
|
+
"name": "AIME 2025",
|
|
1087
|
+
"value": 97.3,
|
|
1088
|
+
"source": "https://huggingface.co/stepfun-ai/Step-3.5-Flash",
|
|
1089
|
+
"asOf": "2026-02-01"
|
|
1090
|
+
},
|
|
1091
|
+
{
|
|
1092
|
+
"name": "SWE-bench Verified",
|
|
1093
|
+
"value": 74.4,
|
|
1094
|
+
"source": "https://huggingface.co/stepfun-ai/Step-3.5-Flash",
|
|
1095
|
+
"asOf": "2026-02-01"
|
|
1096
|
+
}
|
|
1097
|
+
],
|
|
1098
|
+
"docUrl": "https://static.stepfun.com/blog/step-3.5-flash/",
|
|
1099
|
+
"notes": "Step 3.5 Flash uses 288 routed experts per layer plus 1 shared expert, top-8 per token."
|
|
1100
|
+
}
|
|
605
1101
|
},
|
|
606
1102
|
{
|
|
607
1103
|
"id": "openrouter/tencent/hy3",
|
|
608
1104
|
"name": "Tencent HY3 (OR)",
|
|
609
1105
|
"provider": "openrouter",
|
|
610
1106
|
"enabled": true,
|
|
611
|
-
"modality": "text->text"
|
|
1107
|
+
"modality": "text->text",
|
|
1108
|
+
"info": {
|
|
1109
|
+
"params": "Hy3 is a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters and 3.8B MTP layer parameters, developed by the Tencent Hy Team.",
|
|
1110
|
+
"docUrl": "https://github.com/Tencent-Hunyuan/Hy3",
|
|
1111
|
+
"notes": "Hy3 has 192 experts with top-8 activation."
|
|
1112
|
+
}
|
|
612
1113
|
},
|
|
613
1114
|
{
|
|
614
1115
|
"id": "openrouter/tencent/hy3-preview",
|
|
615
1116
|
"name": "Tencent HY3 Preview (OR)",
|
|
616
1117
|
"provider": "openrouter",
|
|
617
1118
|
"enabled": true,
|
|
618
|
-
"modality": "text->text"
|
|
1119
|
+
"modality": "text->text",
|
|
1120
|
+
"info": {
|
|
1121
|
+
"params": "Hy3 preview is a 295B-parameter Mixture-of-Experts (MoE) model with 21B active parameters and 3.8B MTP layer parameters, developed by the Tencent Hy Team.",
|
|
1122
|
+
"docUrl": "https://github.com/Tencent-Hunyuan/Hy3-preview",
|
|
1123
|
+
"notes": "Hy3 preview has 192 experts with top-8 activation."
|
|
1124
|
+
}
|
|
619
1125
|
},
|
|
620
1126
|
{
|
|
621
1127
|
"id": "openrouter/xiaomi/mimo-v2.5-pro",
|
|
622
1128
|
"name": "Xiaomi MiMo V2.5 Pro (OR)",
|
|
623
1129
|
"provider": "openrouter",
|
|
624
1130
|
"enabled": true,
|
|
625
|
-
"modality": "text->text"
|
|
1131
|
+
"modality": "text->text",
|
|
1132
|
+
"info": {
|
|
1133
|
+
"params": "MiMo-V2.5-Pro is an open-source Mixture-of-Experts (MoE) language model with 1.02T total parameters and 42B active parameters.",
|
|
1134
|
+
"docUrl": "https://mimo.xiaomi.com/mimo-v2-5-pro",
|
|
1135
|
+
"notes": "MiMo-V2.5-Pro uses hybrid Sliding Window Attention and Global Attention with a 6:1 ratio."
|
|
1136
|
+
}
|
|
626
1137
|
},
|
|
627
1138
|
{
|
|
628
1139
|
"id": "openrouter/xiaomi/mimo-v2.5",
|
|
629
1140
|
"name": "Xiaomi MiMo V2.5 (OR)",
|
|
630
1141
|
"provider": "openrouter",
|
|
631
1142
|
"enabled": true,
|
|
632
|
-
"modality": "text+image+audio+video->text"
|
|
1143
|
+
"modality": "text+image+audio+video->text",
|
|
1144
|
+
"info": {
|
|
1145
|
+
"params": "Architecture: Sparse MoE (Mixture of Experts), 310B total / 15B activated parameters",
|
|
1146
|
+
"docUrl": "https://mimo.xiaomi.com/mimo-v2-5",
|
|
1147
|
+
"notes": "MiMo-V2.5 is a native omnimodal model supporting text, image, video, and audio understanding."
|
|
1148
|
+
}
|
|
633
1149
|
},
|
|
634
1150
|
{
|
|
635
1151
|
"id": "openrouter/meituan/longcat-2.0",
|
|
636
1152
|
"name": "Meituan LongCat 2.0 (OR)",
|
|
637
1153
|
"provider": "openrouter",
|
|
638
1154
|
"enabled": true,
|
|
639
|
-
"modality": "text->text"
|
|
1155
|
+
"modality": "text->text",
|
|
1156
|
+
"info": {
|
|
1157
|
+
"params": "We introduce LongCat-2.0, a large-scale MoE language model with 1.6 trillion total parameters and ~48 billion activated per token",
|
|
1158
|
+
"benchmarks": [
|
|
1159
|
+
{
|
|
1160
|
+
"name": "Terminal-Bench 2.1",
|
|
1161
|
+
"value": 70.8,
|
|
1162
|
+
"source": "https://huggingface.co/meituan-longcat/LongCat-2.0",
|
|
1163
|
+
"asOf": "2026-07-05"
|
|
1164
|
+
},
|
|
1165
|
+
{
|
|
1166
|
+
"name": "SWE-bench Pro",
|
|
1167
|
+
"value": 59.5,
|
|
1168
|
+
"source": "https://huggingface.co/meituan-longcat/LongCat-2.0",
|
|
1169
|
+
"asOf": "2026-07-05"
|
|
1170
|
+
},
|
|
1171
|
+
{
|
|
1172
|
+
"name": "GPQA-diamond",
|
|
1173
|
+
"value": 88.9,
|
|
1174
|
+
"source": "https://huggingface.co/meituan-longcat/LongCat-2.0",
|
|
1175
|
+
"asOf": "2026-07-05"
|
|
1176
|
+
}
|
|
1177
|
+
],
|
|
1178
|
+
"docUrl": "https://longcat.chat/blog/longcat-2.0",
|
|
1179
|
+
"notes": "LongCat-2.0 includes 135B N-gram Embedding parameters in addition to its MoE."
|
|
1180
|
+
}
|
|
640
1181
|
},
|
|
641
1182
|
{
|
|
642
1183
|
"id": "openrouter/kwaipilot/kat-coder-pro-v2.5",
|
|
@@ -738,28 +1279,107 @@
|
|
|
738
1279
|
"name": "NVIDIA Nemotron 3 Super 120B (OR)",
|
|
739
1280
|
"provider": "openrouter",
|
|
740
1281
|
"enabled": true,
|
|
741
|
-
"modality": "text->text"
|
|
1282
|
+
"modality": "text->text",
|
|
1283
|
+
"info": {
|
|
1284
|
+
"params": "120B total · 12B active",
|
|
1285
|
+
"benchmarks": [
|
|
1286
|
+
{
|
|
1287
|
+
"name": "MMLU-Pro",
|
|
1288
|
+
"value": 83.73,
|
|
1289
|
+
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8",
|
|
1290
|
+
"asOf": "2026-03-11"
|
|
1291
|
+
},
|
|
1292
|
+
{
|
|
1293
|
+
"name": "GPQA (no tools)",
|
|
1294
|
+
"value": 79.23,
|
|
1295
|
+
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8",
|
|
1296
|
+
"asOf": "2026-03-11"
|
|
1297
|
+
},
|
|
1298
|
+
{
|
|
1299
|
+
"name": "HMMT Feb25 (with tools)",
|
|
1300
|
+
"value": 94.73,
|
|
1301
|
+
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8",
|
|
1302
|
+
"asOf": "2026-03-11"
|
|
1303
|
+
}
|
|
1304
|
+
],
|
|
1305
|
+
"docUrl": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8",
|
|
1306
|
+
"notes": "Hybrid LatentMoE architecture interleaving Mamba-2, MoE, and Attention layers with Multi-Token Prediction (MTP) layers."
|
|
1307
|
+
}
|
|
742
1308
|
},
|
|
743
1309
|
{
|
|
744
1310
|
"id": "openrouter/nvidia/nemotron-3-ultra-550b-a55b",
|
|
745
1311
|
"name": "NVIDIA Nemotron 3 Ultra 550B (OR)",
|
|
746
1312
|
"provider": "openrouter",
|
|
747
1313
|
"enabled": true,
|
|
748
|
-
"modality": "text->text"
|
|
1314
|
+
"modality": "text->text",
|
|
1315
|
+
"info": {
|
|
1316
|
+
"params": "550B total · 55B active",
|
|
1317
|
+
"benchmarks": [
|
|
1318
|
+
{
|
|
1319
|
+
"name": "Terminal Bench 2.1",
|
|
1320
|
+
"value": 56.4,
|
|
1321
|
+
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
|
|
1322
|
+
"asOf": "2026-06-04"
|
|
1323
|
+
},
|
|
1324
|
+
{
|
|
1325
|
+
"name": "SWE-Bench Verified",
|
|
1326
|
+
"value": 70.7,
|
|
1327
|
+
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
|
|
1328
|
+
"asOf": "2026-06-04"
|
|
1329
|
+
},
|
|
1330
|
+
{
|
|
1331
|
+
"name": "MMLU-Pro",
|
|
1332
|
+
"value": 86.8,
|
|
1333
|
+
"source": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
|
|
1334
|
+
"asOf": "2026-06-04"
|
|
1335
|
+
}
|
|
1336
|
+
],
|
|
1337
|
+
"docUrl": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
|
|
1338
|
+
"notes": "Hybrid LatentMoE architecture interleaving Mamba-2, MoE, and Attention layers with Multi-Token Prediction (MTP) layers."
|
|
1339
|
+
}
|
|
749
1340
|
},
|
|
750
1341
|
{
|
|
751
1342
|
"id": "openrouter/nousresearch/hermes-4-405b",
|
|
752
1343
|
"name": "Nous Hermes 4 405B (OR)",
|
|
753
1344
|
"provider": "openrouter",
|
|
754
1345
|
"enabled": true,
|
|
755
|
-
"modality": "text->text"
|
|
1346
|
+
"modality": "text->text",
|
|
1347
|
+
"info": {
|
|
1348
|
+
"docUrl": "https://arxiv.org/abs/2508.18255",
|
|
1349
|
+
"notes": "Hybrid-mode reasoning model based on Llama-3.1-405B that emits explicit <think>…</think> segments when it chooses to deliberate."
|
|
1350
|
+
}
|
|
756
1351
|
},
|
|
757
1352
|
{
|
|
758
1353
|
"id": "openrouter/thinkingmachines/inkling",
|
|
759
1354
|
"name": "Thinking Machines Inkling (OR)",
|
|
760
1355
|
"provider": "openrouter",
|
|
761
1356
|
"enabled": true,
|
|
762
|
-
"modality": "text+image+audio->text"
|
|
1357
|
+
"modality": "text+image+audio->text",
|
|
1358
|
+
"info": {
|
|
1359
|
+
"params": "975B total · 41B active",
|
|
1360
|
+
"benchmarks": [
|
|
1361
|
+
{
|
|
1362
|
+
"name": "GPQA Diamond",
|
|
1363
|
+
"value": 87.2,
|
|
1364
|
+
"source": "https://huggingface.co/thinkingmachines/Inkling",
|
|
1365
|
+
"asOf": "2026-07-14"
|
|
1366
|
+
},
|
|
1367
|
+
{
|
|
1368
|
+
"name": "AIME 2026",
|
|
1369
|
+
"value": 97.1,
|
|
1370
|
+
"source": "https://huggingface.co/thinkingmachines/Inkling",
|
|
1371
|
+
"asOf": "2026-07-14"
|
|
1372
|
+
},
|
|
1373
|
+
{
|
|
1374
|
+
"name": "SWEBench Verified",
|
|
1375
|
+
"value": 77.6,
|
|
1376
|
+
"source": "https://huggingface.co/thinkingmachines/Inkling",
|
|
1377
|
+
"asOf": "2026-07-14"
|
|
1378
|
+
}
|
|
1379
|
+
],
|
|
1380
|
+
"docUrl": "https://github.com/thinking-machines-lab/tinker-cookbook",
|
|
1381
|
+
"notes": "Native multimodal MoE accepting text, image and audio input, with images and video encoded via a hierarchical patch encoder and audio via discrete token encoding."
|
|
1382
|
+
}
|
|
763
1383
|
},
|
|
764
1384
|
{
|
|
765
1385
|
"id": "openrouter/relace/relace-apply-3",
|
|
@@ -962,7 +1582,31 @@
|
|
|
962
1582
|
"provider": "openrouter",
|
|
963
1583
|
"enabled": false,
|
|
964
1584
|
"modality": "text+image->text",
|
|
965
|
-
"deprecated": true
|
|
1585
|
+
"deprecated": true,
|
|
1586
|
+
"info": {
|
|
1587
|
+
"benchmarks": [
|
|
1588
|
+
{
|
|
1589
|
+
"name": "Terminal Bench 2.1",
|
|
1590
|
+
"value": 83.9,
|
|
1591
|
+
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp",
|
|
1592
|
+
"asOf": "2026-08-31"
|
|
1593
|
+
},
|
|
1594
|
+
{
|
|
1595
|
+
"name": "DeepSWE",
|
|
1596
|
+
"value": 59.3,
|
|
1597
|
+
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp",
|
|
1598
|
+
"asOf": "2026-08-31"
|
|
1599
|
+
},
|
|
1600
|
+
{
|
|
1601
|
+
"name": "Toolathlon-Verified",
|
|
1602
|
+
"value": 75.9,
|
|
1603
|
+
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp",
|
|
1604
|
+
"asOf": "2026-08-31"
|
|
1605
|
+
}
|
|
1606
|
+
],
|
|
1607
|
+
"docUrl": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-Vision-Exp",
|
|
1608
|
+
"notes": "Experimental multimodal model built on the DeepSeek-V4-Flash architecture, adding visual modules to unlock vision understanding."
|
|
1609
|
+
}
|
|
966
1610
|
},
|
|
967
1611
|
{
|
|
968
1612
|
"id": "openrouter/openai/gpt-5-image",
|
|
@@ -1214,35 +1858,160 @@
|
|
|
1214
1858
|
"name": "Qwen3.5-122B-A10B (OR)",
|
|
1215
1859
|
"provider": "openrouter",
|
|
1216
1860
|
"enabled": false,
|
|
1217
|
-
"modality": "text+image+video->text"
|
|
1861
|
+
"modality": "text+image+video->text",
|
|
1862
|
+
"info": {
|
|
1863
|
+
"params": "122B total · 10B active",
|
|
1864
|
+
"benchmarks": [
|
|
1865
|
+
{
|
|
1866
|
+
"name": "MMLU-Pro",
|
|
1867
|
+
"value": 86.7,
|
|
1868
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B",
|
|
1869
|
+
"asOf": "2026-02-24"
|
|
1870
|
+
},
|
|
1871
|
+
{
|
|
1872
|
+
"name": "GPQA Diamond",
|
|
1873
|
+
"value": 86.6,
|
|
1874
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B",
|
|
1875
|
+
"asOf": "2026-02-24"
|
|
1876
|
+
},
|
|
1877
|
+
{
|
|
1878
|
+
"name": "SWE-bench Verified",
|
|
1879
|
+
"value": 72.0,
|
|
1880
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B",
|
|
1881
|
+
"asOf": "2026-02-24"
|
|
1882
|
+
}
|
|
1883
|
+
],
|
|
1884
|
+
"docUrl": "https://qwen.ai/blog?id=qwen3.5",
|
|
1885
|
+
"notes": "Native vision-language model using Gated Delta Networks plus sparse MoE (8 routed + 1 shared expert per token)."
|
|
1886
|
+
}
|
|
1218
1887
|
},
|
|
1219
1888
|
{
|
|
1220
1889
|
"id": "openrouter/qwen/qwen3.5-27b",
|
|
1221
1890
|
"name": "Qwen3.5-27B (OR)",
|
|
1222
1891
|
"provider": "openrouter",
|
|
1223
1892
|
"enabled": false,
|
|
1224
|
-
"modality": "text+image+video->text"
|
|
1893
|
+
"modality": "text+image+video->text",
|
|
1894
|
+
"info": {
|
|
1895
|
+
"params": "27B parameters",
|
|
1896
|
+
"benchmarks": [
|
|
1897
|
+
{
|
|
1898
|
+
"name": "MMLU-Pro",
|
|
1899
|
+
"value": 86.1,
|
|
1900
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-27B",
|
|
1901
|
+
"asOf": "2026-02-24"
|
|
1902
|
+
},
|
|
1903
|
+
{
|
|
1904
|
+
"name": "GPQA Diamond",
|
|
1905
|
+
"value": 85.5,
|
|
1906
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-27B",
|
|
1907
|
+
"asOf": "2026-02-24"
|
|
1908
|
+
},
|
|
1909
|
+
{
|
|
1910
|
+
"name": "SWE-bench Verified",
|
|
1911
|
+
"value": 72.4,
|
|
1912
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-27B",
|
|
1913
|
+
"asOf": "2026-02-24"
|
|
1914
|
+
}
|
|
1915
|
+
],
|
|
1916
|
+
"docUrl": "https://qwen.ai/blog?id=qwen3.5",
|
|
1917
|
+
"notes": "Native vision-language dense model incorporating a linear attention (Gated DeltaNet) mechanism."
|
|
1918
|
+
}
|
|
1225
1919
|
},
|
|
1226
1920
|
{
|
|
1227
1921
|
"id": "openrouter/qwen/qwen3.5-35b-a3b",
|
|
1228
1922
|
"name": "Qwen3.5-35B-A3B (OR)",
|
|
1229
1923
|
"provider": "openrouter",
|
|
1230
1924
|
"enabled": false,
|
|
1231
|
-
"modality": "text+image+video->text"
|
|
1925
|
+
"modality": "text+image+video->text",
|
|
1926
|
+
"info": {
|
|
1927
|
+
"params": "35B total · 3B active",
|
|
1928
|
+
"benchmarks": [
|
|
1929
|
+
{
|
|
1930
|
+
"name": "MMLU-Pro",
|
|
1931
|
+
"value": 85.3,
|
|
1932
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B",
|
|
1933
|
+
"asOf": "2026-02-24"
|
|
1934
|
+
},
|
|
1935
|
+
{
|
|
1936
|
+
"name": "GPQA Diamond",
|
|
1937
|
+
"value": 84.2,
|
|
1938
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B",
|
|
1939
|
+
"asOf": "2026-02-24"
|
|
1940
|
+
},
|
|
1941
|
+
{
|
|
1942
|
+
"name": "SWE-bench Verified",
|
|
1943
|
+
"value": 69.2,
|
|
1944
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-35B-A3B",
|
|
1945
|
+
"asOf": "2026-02-24"
|
|
1946
|
+
}
|
|
1947
|
+
],
|
|
1948
|
+
"docUrl": "https://qwen.ai/blog?id=qwen3.5",
|
|
1949
|
+
"notes": "Native vision-language model using Gated Delta Networks plus sparse Mixture-of-Experts."
|
|
1950
|
+
}
|
|
1232
1951
|
},
|
|
1233
1952
|
{
|
|
1234
1953
|
"id": "openrouter/qwen/qwen3.5-397b-a17b",
|
|
1235
1954
|
"name": "Qwen3.5 397B A17B (OR)",
|
|
1236
1955
|
"provider": "openrouter",
|
|
1237
1956
|
"enabled": false,
|
|
1238
|
-
"modality": "text+image+video->text"
|
|
1957
|
+
"modality": "text+image+video->text",
|
|
1958
|
+
"info": {
|
|
1959
|
+
"params": "397B in total and 17B activated",
|
|
1960
|
+
"benchmarks": [
|
|
1961
|
+
{
|
|
1962
|
+
"name": "MMLU-Pro",
|
|
1963
|
+
"value": 87.8,
|
|
1964
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B",
|
|
1965
|
+
"asOf": "2026-02-16"
|
|
1966
|
+
},
|
|
1967
|
+
{
|
|
1968
|
+
"name": "GPQA",
|
|
1969
|
+
"value": 88.4,
|
|
1970
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B",
|
|
1971
|
+
"asOf": "2026-02-16"
|
|
1972
|
+
},
|
|
1973
|
+
{
|
|
1974
|
+
"name": "SWE-bench Verified",
|
|
1975
|
+
"value": 76.4,
|
|
1976
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-397B-A17B",
|
|
1977
|
+
"asOf": "2026-02-16"
|
|
1978
|
+
}
|
|
1979
|
+
],
|
|
1980
|
+
"docUrl": "https://github.com/QwenLM/Qwen3.8",
|
|
1981
|
+
"notes": "Qwen3.5-397B-A17B uses a Gated DeltaNet plus sparse Mixture-of-Experts architecture with 512 experts (10 routed + 1 shared)."
|
|
1982
|
+
}
|
|
1239
1983
|
},
|
|
1240
1984
|
{
|
|
1241
1985
|
"id": "openrouter/qwen/qwen3.5-9b",
|
|
1242
1986
|
"name": "Qwen3.5-9B (OR)",
|
|
1243
1987
|
"provider": "openrouter",
|
|
1244
1988
|
"enabled": false,
|
|
1245
|
-
"modality": "text+image+video->text"
|
|
1989
|
+
"modality": "text+image+video->text",
|
|
1990
|
+
"info": {
|
|
1991
|
+
"params": "9B parameters",
|
|
1992
|
+
"benchmarks": [
|
|
1993
|
+
{
|
|
1994
|
+
"name": "MMLU-Pro",
|
|
1995
|
+
"value": 82.5,
|
|
1996
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-9B",
|
|
1997
|
+
"asOf": "2026-02-27"
|
|
1998
|
+
},
|
|
1999
|
+
{
|
|
2000
|
+
"name": "GPQA Diamond",
|
|
2001
|
+
"value": 81.7,
|
|
2002
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-9B",
|
|
2003
|
+
"asOf": "2026-02-27"
|
|
2004
|
+
},
|
|
2005
|
+
{
|
|
2006
|
+
"name": "LiveCodeBench v6",
|
|
2007
|
+
"value": 65.6,
|
|
2008
|
+
"source": "https://huggingface.co/Qwen/Qwen3.5-9B",
|
|
2009
|
+
"asOf": "2026-02-27"
|
|
2010
|
+
}
|
|
2011
|
+
],
|
|
2012
|
+
"docUrl": "https://qwen.ai/blog?id=qwen3.5",
|
|
2013
|
+
"notes": "Native vision-language foundation model with a unified vision-language design."
|
|
2014
|
+
}
|
|
1246
2015
|
},
|
|
1247
2016
|
{
|
|
1248
2017
|
"id": "openrouter/qwen/qwen3.5-flash-02-23",
|
|
@@ -1281,14 +2050,64 @@
|
|
|
1281
2050
|
"name": "Qwen3.8 27B (OR)",
|
|
1282
2051
|
"provider": "openrouter",
|
|
1283
2052
|
"enabled": false,
|
|
1284
|
-
"modality": "text+image+video->text"
|
|
2053
|
+
"modality": "text+image+video->text",
|
|
2054
|
+
"info": {
|
|
2055
|
+
"params": "27B parameters",
|
|
2056
|
+
"benchmarks": [
|
|
2057
|
+
{
|
|
2058
|
+
"name": "SWE-bench Pro",
|
|
2059
|
+
"value": 61.7,
|
|
2060
|
+
"source": "https://huggingface.co/Qwen/Qwen3.8-27B",
|
|
2061
|
+
"asOf": "2026-08-05"
|
|
2062
|
+
},
|
|
2063
|
+
{
|
|
2064
|
+
"name": "GPQA Diamond",
|
|
2065
|
+
"value": 89.2,
|
|
2066
|
+
"source": "https://huggingface.co/Qwen/Qwen3.8-27B",
|
|
2067
|
+
"asOf": "2026-08-05"
|
|
2068
|
+
},
|
|
2069
|
+
{
|
|
2070
|
+
"name": "LiveCodeBench v6",
|
|
2071
|
+
"value": 90.3,
|
|
2072
|
+
"source": "https://huggingface.co/Qwen/Qwen3.8-27B",
|
|
2073
|
+
"asOf": "2026-08-05"
|
|
2074
|
+
}
|
|
2075
|
+
],
|
|
2076
|
+
"docUrl": "https://www.qwencloud.com/models/qwen3.8-27b",
|
|
2077
|
+
"notes": "Qwen3.8-27B is a native vision-language model with flexible thinking control via reasoning_effort and preserve_thinking."
|
|
2078
|
+
}
|
|
1285
2079
|
},
|
|
1286
2080
|
{
|
|
1287
2081
|
"id": "openrouter/qwen/qwen3.8-flash",
|
|
1288
2082
|
"name": "Qwen3.8 Flash (OR)",
|
|
1289
2083
|
"provider": "openrouter",
|
|
1290
2084
|
"enabled": false,
|
|
1291
|
-
"modality": "text+image+video->text"
|
|
2085
|
+
"modality": "text+image+video->text",
|
|
2086
|
+
"info": {
|
|
2087
|
+
"params": "125B with 6B activated, plus 51B n-gram embedding and 4B MTP",
|
|
2088
|
+
"benchmarks": [
|
|
2089
|
+
{
|
|
2090
|
+
"name": "GPQA Diamond",
|
|
2091
|
+
"value": 91.7,
|
|
2092
|
+
"source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
|
|
2093
|
+
"asOf": "2026-08-24"
|
|
2094
|
+
},
|
|
2095
|
+
{
|
|
2096
|
+
"name": "LiveCodeBench v6",
|
|
2097
|
+
"value": 91.9,
|
|
2098
|
+
"source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
|
|
2099
|
+
"asOf": "2026-08-24"
|
|
2100
|
+
},
|
|
2101
|
+
{
|
|
2102
|
+
"name": "DeepSWE 1.1",
|
|
2103
|
+
"value": 58.7,
|
|
2104
|
+
"source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
|
|
2105
|
+
"asOf": "2026-08-24"
|
|
2106
|
+
}
|
|
2107
|
+
],
|
|
2108
|
+
"docUrl": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
|
|
2109
|
+
"notes": "Uses an n-gram embedding for parameter scaling alongside a sparse Mixture-of-Experts backbone."
|
|
2110
|
+
}
|
|
1292
2111
|
},
|
|
1293
2112
|
{
|
|
1294
2113
|
"id": "openrouter/qwen/qwen3.8-max-0902",
|
|
@@ -1309,21 +2128,94 @@
|
|
|
1309
2128
|
"name": "DeepSeek V4 Flash 0731 (OR)",
|
|
1310
2129
|
"provider": "openrouter",
|
|
1311
2130
|
"enabled": false,
|
|
1312
|
-
"modality": "text->text"
|
|
2131
|
+
"modality": "text->text",
|
|
2132
|
+
"info": {
|
|
2133
|
+
"benchmarks": [
|
|
2134
|
+
{
|
|
2135
|
+
"name": "Terminal Bench 2.1",
|
|
2136
|
+
"value": 82.7,
|
|
2137
|
+
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731",
|
|
2138
|
+
"asOf": "2026-07-31"
|
|
2139
|
+
},
|
|
2140
|
+
{
|
|
2141
|
+
"name": "DeepSWE",
|
|
2142
|
+
"value": 54.4,
|
|
2143
|
+
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731",
|
|
2144
|
+
"asOf": "2026-07-31"
|
|
2145
|
+
},
|
|
2146
|
+
{
|
|
2147
|
+
"name": "Toolathlon-Verified",
|
|
2148
|
+
"value": 70.3,
|
|
2149
|
+
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731",
|
|
2150
|
+
"asOf": "2026-07-31"
|
|
2151
|
+
}
|
|
2152
|
+
],
|
|
2153
|
+
"docUrl": "https://arxiv.org/abs/2606.19348",
|
|
2154
|
+
"notes": "Official DeepSeek-V4-Flash release superseding the preview, with an attached DSpark speculative decoding module."
|
|
2155
|
+
}
|
|
1313
2156
|
},
|
|
1314
2157
|
{
|
|
1315
2158
|
"id": "openrouter/deepseek/deepseek-v4-pro-0813",
|
|
1316
2159
|
"name": "DeepSeek V4 Pro 0813 (OR)",
|
|
1317
2160
|
"provider": "openrouter",
|
|
1318
2161
|
"enabled": false,
|
|
1319
|
-
"modality": "text->text"
|
|
2162
|
+
"modality": "text->text",
|
|
2163
|
+
"info": {
|
|
2164
|
+
"benchmarks": [
|
|
2165
|
+
{
|
|
2166
|
+
"name": "Terminal Bench 2.1",
|
|
2167
|
+
"value": 87.9,
|
|
2168
|
+
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813",
|
|
2169
|
+
"asOf": "2026-08-13"
|
|
2170
|
+
},
|
|
2171
|
+
{
|
|
2172
|
+
"name": "DeepSWE",
|
|
2173
|
+
"value": 62.7,
|
|
2174
|
+
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813",
|
|
2175
|
+
"asOf": "2026-08-13"
|
|
2176
|
+
},
|
|
2177
|
+
{
|
|
2178
|
+
"name": "Toolathlon-Verified",
|
|
2179
|
+
"value": 74.1,
|
|
2180
|
+
"source": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro-0813",
|
|
2181
|
+
"asOf": "2026-08-13"
|
|
2182
|
+
}
|
|
2183
|
+
],
|
|
2184
|
+
"docUrl": "https://arxiv.org/abs/2606.19348",
|
|
2185
|
+
"notes": "Official DeepSeek-V4-Pro release superseding the preview, with an attached DSpark speculative decoding module."
|
|
2186
|
+
}
|
|
1320
2187
|
},
|
|
1321
2188
|
{
|
|
1322
2189
|
"id": "openrouter/moonshotai/kimi-k2.5",
|
|
1323
2190
|
"name": "Kimi K2.5 (OR)",
|
|
1324
2191
|
"provider": "openrouter",
|
|
1325
2192
|
"enabled": false,
|
|
1326
|
-
"modality": "text+image->text"
|
|
2193
|
+
"modality": "text+image->text",
|
|
2194
|
+
"info": {
|
|
2195
|
+
"params": "1T total parameters · 32B activated parameters",
|
|
2196
|
+
"benchmarks": [
|
|
2197
|
+
{
|
|
2198
|
+
"name": "GPQA-Diamond",
|
|
2199
|
+
"value": 87.6,
|
|
2200
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.5",
|
|
2201
|
+
"asOf": "2026-01-29"
|
|
2202
|
+
},
|
|
2203
|
+
{
|
|
2204
|
+
"name": "SWE-Bench Verified",
|
|
2205
|
+
"value": 76.8,
|
|
2206
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.5",
|
|
2207
|
+
"asOf": "2026-01-29"
|
|
2208
|
+
},
|
|
2209
|
+
{
|
|
2210
|
+
"name": "AIME 2025",
|
|
2211
|
+
"value": 96.1,
|
|
2212
|
+
"source": "https://huggingface.co/moonshotai/Kimi-K2.5",
|
|
2213
|
+
"asOf": "2026-01-29"
|
|
2214
|
+
}
|
|
2215
|
+
],
|
|
2216
|
+
"docUrl": "https://platform.moonshot.ai",
|
|
2217
|
+
"notes": "Native multimodal agentic model with a MoonViT vision encoder (400M parameters) and shared experts."
|
|
2218
|
+
}
|
|
1327
2219
|
},
|
|
1328
2220
|
{
|
|
1329
2221
|
"id": "openrouter/x-ai/grok-4.6",
|