🏆 Rankings
Highscore
The best runs in a direct ranking – switchable between performance (tokens/s) and harness (task points).
| # | Model / Maker | Graphics card | CPU / RAM | Generation | Prefill | TG | TTFT | Parallel | Runtime | |
|---|---|---|---|---|---|---|---|---|---|---|
| 🥇 | NVIDIA-Nemotron-3.5-Lightning-30B-A3B31.6BNVIDIA | NVIDIA GeForce RTX 3090 | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 3.484,3 tok/s |
8.256 | 3.484,3 | 14387 ms | 10× | tensorfoldgodclaw 1.1.6MLX 4-bit | Details → |
| 🥈 | NVIDIA-Nemotron-3.5-Lightning-30B-A3B31.6BNVIDIA | NVIDIA GeForce RTX 3090 | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 2.867,2 tok/s |
8.107 | 2.867,2 | 17054 ms | 10× | tensorfoldgodclaw 1.1.6MLX 4-bit | Details → |
| 🥉 | gemma-4-E2B-it5BGoogle | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 2.491,2 tok/s |
15.814 | 2.491,2 | 4382 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 4 | gemma-4-E2B-it5BGoogle | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 2.414,4 tok/s |
20.277 | 2.414,4 | 4106 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 5 | gemma-4-E2B-it5BGoogle | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 2.412,3 tok/s |
19.677 | 2.412,3 | 4106 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 6 | NVIDIA-Nemotron-3.5-Lightning-30B-A3B31.6BNVIDIA | NVIDIA GeForce RTX 3090 Ti | AMD Ryzen 5 5600X 6-Core Processor 30 GB | 2.263,7 tok/s |
9.016 | 2.263,7 | 21817 ms | 10× | vLLMvllmNVFP4 | Details → |
| 7 | NVIDIA-Nemotron-3-Nano-4B4BNVIDIA | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 2.182,7 tok/s |
10.599 | 2.182,7 | 5597 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 8 | NVIDIA-Nemotron-3-Nano-4B4BNVIDIA | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 2.144,3 tok/s |
10.533 | 2.144,3 | 5597 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 9 | gemma-4-E2B-it5BGoogle | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 2.143,6 tok/s |
13.632 | 2.143,6 | 4819 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 10 | gpt-oss-20b20BOpenAI | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.995,4 tok/s |
16.158 | 1.995,4 | 5309 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 11 | gpt-oss-20b20BOpenAI | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.993,5 tok/s |
16.318 | 1.993,5 | 5322 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 12 | gpt-oss-20b20BOpenAI | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 1.981,1 tok/s |
15.983 | 1.981,1 | 5309 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 13 | NVIDIA-Nemotron-3-Nano-4B4BNVIDIA | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 1.937,7 tok/s |
9.452 | 1.937,7 | 6287 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 14 | NVIDIA-Nemotron-3-Nano-4B4BNVIDIA | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 1.911,6 tok/s |
10.116 | 1.911,6 | 6244 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 15 | NVIDIA-Nemotron-3-Nano-4B4BNVIDIA | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 1.897,0 tok/s |
10.984 | 1.897,0 | 6226 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 16 | gpt-oss-20b20BOpenAI | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.890,7 tok/s |
14.771 | 1.890,7 | 5489 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 17 | NVIDIA-Nemotron-3-Nano-4B4BNVIDIA | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.878,7 tok/s |
7.717 | 1.878,7 | 6689 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 18 | gpt-oss-20b20BOpenAI | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 1.835,7 tok/s |
15.972 | 1.835,7 | 5760 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 19 | NVIDIA-Nemotron-3-Nano-4B4BNVIDIA | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 1.830,3 tok/s |
9.399 | 1.830,3 | 6598 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 20 | gpt-oss-20b20BOpenAI | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 1.819,5 tok/s |
19.732 | 1.819,5 | 5578 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 21 | gpt-oss-20b20BOpenAI | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 1.802,0 tok/s |
18.226 | 1.802,0 | 5604 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 22 | Ornith-1.0-35B35BDeepReinforce | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 1.799,4 tok/s |
9.192 | 1.799,4 | 8876 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 23 | Nemotron-3-Nano-Omni-30B-A3B-Reasoning30BNVIDIA | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.791,5 tok/s |
7.972 | 1.791,5 | 6996 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 24 | Nemotron-Cascade-2-30B-A3B30BNVIDIA | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 1.784,7 tok/s |
8.304 | 1.784,7 | 6952 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 25 | NVIDIA-Nemotron-3-Nano-30B-A3B30BNVIDIA | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 1.775,6 tok/s |
8.642 | 1.775,6 | 6861 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 26 | Nemotron-3-Nano-Omni-30B-A3B-Reasoning30BNVIDIA | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 1.774,4 tok/s |
7.850 | 1.774,4 | 6996 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 27 | NVIDIA-Nemotron-3-Nano-30B-A3B30BNVIDIA | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.773,5 tok/s |
8.720 | 1.773,5 | 6861 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 28 | Nemotron-Cascade-2-30B-A3B30BNVIDIA | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.768,1 tok/s |
8.360 | 1.768,1 | 6952 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 29 | Laguna-XS-2.133.4BPoolside | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.765,0 tok/s |
8.542 | 1.765,0 | 7776 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 30 | gemma-4-E4B-it8BGoogle | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.758,6 tok/s |
13.889 | 1.758,6 | 5964 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 31 | Laguna-XS-2.133.4BPoolside | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.757,4 tok/s |
9.025 | 1.757,4 | 7343 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 32 | gemma-4-E4B-it8BGoogle | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 1.755,5 tok/s |
13.910 | 1.755,5 | 5964 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 33 | Laguna-XS-2.133.4BPoolside | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.753,8 tok/s |
9.267 | 1.753,8 | 7328 ms | 10× | llama.cppgodclawINT4 | Details → |
| 34 | gemma-4-E2B-it5BGoogle | NVIDIA GeForce RTX 5070 Ti | AMD Ryzen Threadripper PRO 5975WX 32-Cores 247 GB | 1.744,0 tok/s |
13.641 | 1.744,0 | 5790 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 35 | gemma-4-E2B-it5BGoogle | NVIDIA GeForce RTX 5070 Ti | AMD Ryzen Threadripper PRO 5975WX 32-Cores 247 GB | 1.743,1 tok/s |
12.148 | 1.743,1 | 5881 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 36 | gpt-oss-20b20BOpenAI | 3x NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | AMD Ryzen Threadripper PRO 9965WX 24-Cores 125 GB | 1.742,1 tok/s |
12.852 | 1.742,1 | 6156 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 37 | Laguna-XS-2.133.4BPoolside | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.740,4 tok/s |
9.653 | 1.740,4 | 7257 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 38 | Laguna-XS-2.133.4BPoolside | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.740,1 tok/s |
8.571 | 1.740,1 | 7785 ms | 10× | llama.cppcodex_cliINT4 | Details → |
| 39 | Laguna-XS-2.133.4BPoolside | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.738,4 tok/s |
8.872 | 1.738,4 | 7721 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 40 | NVIDIA-Nemotron-3-Nano-30B-A3B30BNVIDIA | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.722,1 tok/s |
7.387 | 1.722,1 | 7499 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 41 | Nemotron-Cascade-2-30B-A3B30BNVIDIA | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.719,2 tok/s |
8.044 | 1.719,2 | 7375 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 42 | Qwen3-Coder-30B-A3B-Instruct30BQwen (Alibaba) | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.716,6 tok/s |
12.604 | 1.716,6 | 6861 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 43 | Nemotron-3-Nano-Omni-30B-A3B-Reasoning30BNVIDIA | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.706,7 tok/s |
7.511 | 1.706,7 | 7506 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 44 | Laguna-XS-2.133.4BPoolside | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 1.706,2 tok/s |
8.777 | 1.706,2 | 7343 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 45 | NVIDIA-Nemotron-3.5-Lightning-30B-A3B31.6BNVIDIA | NVIDIA GeForce RTX 3090 | AMD Ryzen 5 5600X 6-Core Processor 31 GB | 1.694,7 tok/s |
7.768 | 1.694,7 | 6482 ms | 5× | tensorfoldgodclaw 1.1.6MLX 4-bit | Details → |
| 46 | Qwen3-Coder-30B-A3B-Instruct30BQwen (Alibaba) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.690,0 tok/s |
14.458 | 1.690,0 | 6608 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
| 47 | Qwen3-30B-A3B-Thinking-250730BQwen (Alibaba) | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.683,8 tok/s |
10.169 | 1.683,8 | 7361 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 48 | Qwen3-Omni-30B-A3B-Thinking30BQwen (Alibaba) | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.673,6 tok/s |
9.186 | 1.673,6 | 7566 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 49 | Qwen3-30B-A3B-Instruct-250730BQwen (Alibaba) | NVIDIA GeForce RTX 5090 | AMD Ryzen 7 5800X3D 8-Core Processor 126 GB | 1.671,0 tok/s |
11.920 | 1.671,0 | 7068 ms | 10× | llama.cppcodex_cliQ4_K_M | Details → |
| 50 | Qwen3-30B-A3B-Thinking-250730BQwen (Alibaba) | NVIDIA RTX PRO 6000 Blackwell Workstation Edition | AMD Ryzen 9 9950X 16-Core Processor 92 GB | 1.670,5 tok/s |
10.926 | 1.670,5 | 7140 ms | 10× | llama.cppgodclawQ4_K_M | Details → |
Ranking from published benchmarks. Sorted by Generation throughput. How is it measured?

