🏆 Rankings
Highscore
The best runs in a direct ranking – switchable between performance (tokens/s) and harness (task points).
| # | Model / Maker | Graphics card | CPU / RAM | Prefill | Prefill | TG | TTFT | Parallel | Runtime | |
|---|---|---|---|---|---|---|---|---|---|---|
| 🥇 | gemma-4-E2B-it5BGoogle | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 55.259,6 tok/s |
55.260 | 1.568,0 | 492 ms | 10× | vLLMgodclawFP8 | Details → |
| 🥈 | gemma-4-E2B-it5BGoogle | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 47.158,4 tok/s |
47.158 | 1.359,1 | 587 ms | 10× | vLLMgodclawFP8 | Details → |
| 🥉 | Nemotron-3-Nano-4B4BNVIDIA | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 44.171,1 tok/s |
44.171 | 1.485,1 | 716 ms | 10× | vLLMgodclawFP8 | Details → |
| 4 | gemma-4-E2B-it5BGoogle | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 43.093,1 tok/s |
43.093 | 909,0 | 287 ms | 5× | vLLMgodclawFP8 | Details → |
| 5 | Laguna-XS-2.1Poolside | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 39.157,1 tok/s |
39.157 | 836,4 | 1032 ms | 10× | vLLMgodclawAWQ | Details → |
| 6 | Qwen3-30B-A3B-Thinking-250730BQwen (Alibaba) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 39.028,9 tok/s |
39.029 | 906,0 | 1192 ms | 10× | vLLMgodclawAWQ | Details → |
| 7 | gemma-4-E2B-it5BGoogle | NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 38.712,2 tok/s |
38.712 | 1.122,5 | 690 ms | 10× | vLLMopenclaw_cliFP8 | Details → |
| 8 | Qwen3-Coder-30B-A3B-Instruct30BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 37.675,6 tok/s |
37.676 | 789,2 | 952 ms | 10× | vLLMgodclawAWQ | Details → |
| 9 | Qwen3-30B-A3B-Instruct-250730BQwen (Alibaba) | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 37.116,1 tok/s |
37.116 | 909,3 | 1783 ms | 10× | vLLMcodex_cliAWQ | Details → |
| 10 | Qwen3-VL-30B-A3B-Instruct30BQwen (Alibaba) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 36.884,1 tok/s |
36.884 | 1.028,5 | 915 ms | 10× | vLLMgodclawAWQ | Details → |
| 11 | Qwen3-VL-30B-A3B-Instruct30BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 35.900,1 tok/s |
35.900 | 763,5 | 986 ms | 10× | vLLMgodclawAWQ | Details → |
| 12 | gemma-4-E2B-it5BGoogle | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 35.757,2 tok/s |
35.757 | 803,2 | 347 ms | 5× | vLLMgodclawFP8 | Details → |
| 13 | Nemotron-3.5-Lightning-30B-A3B30BNVIDIA | 4x AMD Radeon AI PRO R9700 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 35.719,4 tok/s |
35.719 | 290,3 | 1014 ms | 10× | vLLMgodclawBF16 | Details → |
| 14 | gpt-oss-20b20BOpenAI | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 35.347,7 tok/s |
35.348 | 1.443,9 | 560 ms | 10× | vLLMgodclaw | Details → |
| 15 | Laguna-XS-2.1Poolside | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 35.143,3 tok/s |
35.143 | 712,0 | 1137 ms | 10× | vLLMgodclawFP8 | Details → |
| 16 | Qwen3-30B-A3B-Instruct-250730BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 34.912,8 tok/s |
34.913 | 758,7 | 991 ms | 10× | vLLMgodclawAWQ | Details → |
| 17 | Laguna-XS-2.1Poolside | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 34.194,0 tok/s |
34.194 | 771,8 | 1183 ms | 10× | vLLMgodclawAWQ | Details → |
| 18 | Qwen3-Omni-30B-A3B-Thinking30BQwen (Alibaba) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 33.906,7 tok/s |
33.907 | 910,7 | 1527 ms | 10× | vLLMgodclawAWQ | Details → |
| 19 | Qwen3-Coder-30B-A3B-Instruct30BQwen (Alibaba) | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 33.767,8 tok/s |
33.768 | 966,2 | 1792 ms | 10× | vLLMcodex_cliAWQ | Details → |
| 20 | Qwen3-Coder-30B-A3B-Instruct30BQwen (Alibaba) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 33.740,5 tok/s |
33.741 | 1.070,6 | 975 ms | 10× | vLLMgodclawAWQ | Details → |
| 21 | North-Mini-Code-1.0Cohere | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 33.281,4 tok/s |
33.281 | 518,1 | 869 ms | 10× | vLLMgodclawFP8 | Details → |
| 22 | Qwen3-30B-A3B-Thinking-250730BQwen (Alibaba) | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 33.022,6 tok/s |
33.023 | 873,7 | 1834 ms | 10× | vLLMcodex_cliAWQ | Details → |
| 23 | Qwen3.5-35B-A3B35BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 32.522,1 tok/s |
32.522 | 787,3 | 826 ms | 10× | vLLMgodclawAWQ | Details → |
| 24 | Qwen3-30B-A3B-Thinking-250730BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 32.339,6 tok/s |
32.340 | 751,7 | 1270 ms | 10× | vLLMgodclawAWQ | Details → |
| 25 | Qwen3.6-35B-A3B35BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 32.313,4 tok/s |
32.313 | 800,4 | 826 ms | 10× | vLLMgodclawAWQ | Details → |
| 26 | Qwen3-30B-A3B-Instruct-250730BQwen (Alibaba) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 32.304,9 tok/s |
32.305 | 961,5 | 1633 ms | 10× | vLLMgodclawAWQ | Details → |
| 27 | Gemma-4-26B-A4B-it26BGoogle | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 32.027,3 tok/s |
32.027 | 841,8 | 922 ms | 10× | vLLMgodclawAWQ | Details → |
| 28 | gemma-4-E2B-it5BGoogle | NVIDIA GeForce RTX 3090 Ti | AMD Ryzen 5 5600X 6-Core Processor 30 GB | 32.005,2 tok/s |
32.005 | 451,1 | 864 ms | 10× | vLLMvllm | Details → |
| 29 | Qwen-AgentWorld-35B-A3B35BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 31.954,3 tok/s |
31.954 | 799,7 | 833 ms | 10× | vLLMgodclawAWQ | Details → |
| 30 | Qwen3-Omni-30B-A3B-Thinking30BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 31.909,7 tok/s |
31.910 | 515,3 | 1106 ms | 10× | vLLMgodclawFP8 | Details → |
| 31 | Qwen3-30B-A3B-Thinking-250730BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 31.732,8 tok/s |
31.733 | 554,2 | 1123 ms | 10× | vLLMgodclawFP8 | Details → |
| 32 | Qwen3-30B-A3B-Thinking-250730BQwen (Alibaba) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 31.724,4 tok/s |
31.724 | 625,7 | 395 ms | 5× | vLLMgodclawAWQ | Details → |
| 33 | Qwen3-Coder-30B-A3B-Instruct30BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 31.646,6 tok/s |
31.647 | 582,7 | 1123 ms | 10× | vLLMgodclawFP8 | Details → |
| 34 | gemma-4-E2B-it5BGoogle | NVIDIA GeForce RTX 5070 Ti | AMD Ryzen Threadripper PRO 5975WX 32-Cores 247 GB | 31.451,6 tok/s |
31.452 | 1.608,5 | 1632 ms | 10× | vLLMgodclaw | Details → |
| 35 | gpt-oss-20b20BOpenAI | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 31.097,5 tok/s |
31.098 | 842,9 | 318 ms | 5× | vLLMgodclaw | Details → |
| 36 | Qwen3-30B-A3B-Instruct-250730BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 31.043,0 tok/s |
31.043 | 573,6 | 1133 ms | 10× | vLLMgodclawFP8 | Details → |
| 37 | Qwen3-Coder-30B-A3B-Instruct30BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 30.925,8 tok/s |
30.926 | 720,9 | 1138 ms | 10× | vLLMgodclawAWQ | Details → |
| 38 | Qwen3-30B-A3B-Instruct-250730BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 30.804,9 tok/s |
30.805 | 691,3 | 1147 ms | 10× | vLLMgodclawAWQ | Details → |
| 39 | Qwen3-VL-30B-A3B-Instruct30BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 30.727,3 tok/s |
30.727 | 547,7 | 1142 ms | 10× | vLLMgodclawFP8 | Details → |
| 40 | Laguna-XS-2.1Poolside | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 30.670,5 tok/s |
30.671 | 655,4 | 1283 ms | 10× | vLLMgodclawFP8 | Details → |
| 41 | Qwen3-30B-A3B-Thinking-250730BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 30.511,9 tok/s |
30.512 | 673,2 | 1156 ms | 10× | vLLMgodclawAWQ | Details → |
| 42 | Qwen3-VL-30B-A3B-Instruct30BQwen (Alibaba) | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 30.227,2 tok/s |
30.227 | 697,0 | 1163 ms | 10× | vLLMgodclawAWQ | Details → |
| 43 | Nemotron-3-Nano-4B4BNVIDIA | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 30.214,3 tok/s |
30.214 | 899,4 | 472 ms | 5× | vLLMgodclawFP8 | Details → |
| 44 | North-Mini-Code-1.0Cohere | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 30.082,6 tok/s |
30.083 | 598,9 | 947 ms | 10× | vLLMgodclawAWQ | Details → |
| 45 | Laguna-XS-2.1Poolside | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 29.873,2 tok/s |
29.873 | 580,5 | 586 ms | 5× | vLLMgodclawAWQ | Details → |
| 46 | Gemma-4-26B-A4B-it26BGoogle | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 29.512,9 tok/s |
29.513 | 690,9 | 1000 ms | 10× | vLLMgodclawFP8 | Details → |
| 47 | North-Mini-Code-1.0Cohere | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 29.491,0 tok/s |
29.491 | 648,5 | 1309 ms | 10× | vLLMgodclawAWQ | Details → |
| 48 | gemma-4-E2B-it5BGoogle | NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 29.258,6 tok/s |
29.259 | 631,7 | 405 ms | 5× | vLLMopenclaw_cliFP8 | Details → |
| 49 | Qwen3-Omni-30B-A3B-Thinking30BQwen (Alibaba) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 28.789,9 tok/s |
28.790 | 618,2 | 435 ms | 5× | vLLMgodclawAWQ | Details → |
| 50 | Ornith-1.5-35B-A3B35BDeepReinforce | 2x NVIDIA RTX A6000 | AMD Ryzen Threadripper PRO 7955WX 16-Cores 184 GB | 28.681,2 tok/s |
28.681 | 626,4 | 927 ms | 10× | vLLMgodclawFP8 | Details → |
Ranking from published benchmarks. Sorted by Prefill throughput. How is it measured?