⚡
Performance benchmark
How fast is the model?
Raw throughput on real hardware – how many tokens a model generates per second, how quickly it processes the prompt and how short the time to first token is. Higher is better.
⚡ Generation (tok/s)🔄 Prefill⏱️ TTFT👥 Concurrency
| # | Model / Maker | Metrics | Parallel | GPU / CPU / RAM | Runtime | ||
|---|---|---|---|---|---|---|---|
| 1 | Qwen3-0.6B0.6BQwen (Alibaba) Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 495,51 tok/s TG Prefill 19.609 · TTFT 628 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 2 | gemma-4-E2B-it5BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 465,47 tok/s TG Prefill 8.435 · TTFT 2.408 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 3 | gemma-4-E2B-it5BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 455,35 tok/s TG Prefill 8.245 · TTFT 2.462 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 4 | gpt-oss-20b20BOpenAI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 357,72 tok/s TG Prefill 4.799 · TTFT 4.228 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 5 | gpt-oss-20b20BOpenAI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 347,49 tok/s TG Prefill 4.656 · TTFT 4.348 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 6 | gemma-4-E2B-it5BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 331,51 tok/s TG Prefill 6.908 · TTFT 1.433 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 7 | gemma-4-E2B-it5BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 316,40 tok/s TG Prefill 7.121 · TTFT 1.391 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 8 | Gemma-4-26B-A4B-it26BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 315,53 tok/s TG Prefill 2.703 · TTFT 7.505 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 9 | gpt-oss-20b20BOpenAI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 314,82 tok/s TG Prefill 4.224 · TTFT 2.350 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 10 | gpt-oss-20b20BOpenAI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 310,59 tok/s TG Prefill 4.175 · TTFT 2.376 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 11 | Qwen3-0.6B0.6BQwen (Alibaba) Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 243,73 tok/s TG Prefill 13.492 · TTFT 181 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 12 | Laguna-XS-2.1Poolside Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 223,82 tok/s TG Prefill 3.920 · TTFT 3.793 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 13 | gemma-4-12B-it12BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 212,89 tok/s TG Prefill 638 · TTFT 31.124 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 14 | gemma-4-12B-it12BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 209,09 tok/s TG Prefill 646 · TTFT 30.756 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 15 | Gemma-4-26B-A4B-it26BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 207,68 tok/s TG Prefill 2.458 · TTFT 4.035 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 16 | Devstral-Small-250724BMistral AI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 156,79 tok/s TG Prefill 4.188 · TTFT 11.390 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 17 | gpt-oss-20b20BOpenAI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 154,86 tok/s TG Prefill 3.730 · TTFT 530 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 18 | gpt-oss-20b20BOpenAI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 150,51 tok/s TG Prefill 3.701 · TTFT 533 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 19 | Devstral-Small-2-24B-Instruct-251224BMistral AI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 145,92 tok/s TG Prefill 2.977 · TTFT 10.664 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 20 | gemma-4-12B-it12BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 140,99 tok/s TG Prefill 1.453 · TTFT 6.814 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 21 | gemma-4-12B-it12BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 135,10 tok/s TG Prefill 1.505 · TTFT 6.576 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 22 | Codestral-22B-v0.122BMistral AI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 134,10 tok/s TG Prefill 1.552 · TTFT 19.988 ms | 10× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 23 | gemma-4-E2B-it5BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 133,22 tok/s TG Prefill 5.127 · TTFT 390 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 24 | gemma-4-E2B-it5BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 112,54 tok/s TG Prefill 5.735 · TTFT 345 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 25 | Devstral-Small-250724BMistral AI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 107,29 tok/s TG Prefill 2.668 · TTFT 7.432 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 26 | Laguna-XS-2.1Poolside Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 104,95 tok/s TG Prefill 3.020 · TTFT 909 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 27 | Devstral-Small-2-24B-Instruct-251224BMistral AI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 98,13 tok/s TG Prefill 2.200 · TTFT 6.014 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 28 | Codestral-22B-v0.122BMistral AI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 91,65 tok/s TG Prefill 2.058 · TTFT 7.712 ms | 5× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 29 | Gemma-4-26B-A4B-it26BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 82,20 tok/s TG Prefill 2.622 · TTFT 754 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 30 | gemma-4-12B-it12BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 60,38 tok/s TG Prefill 1.447 · TTFT 1.381 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 31 | Devstral-Small-250724BMistral AI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 46,30 tok/s TG Prefill 1.883 · TTFT 1.835 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 32 | gemma-4-12B-it12BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 46,04 tok/s TG Prefill 1.508 · TTFT 1.312 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ8_0 | Details → | |
| 33 | Devstral-Small-2-24B-Instruct-251224BMistral AI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 45,06 tok/s TG Prefill 1.195 · TTFT 1.788 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → | |
| 34 | Codestral-22B-v0.122BMistral AI Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 44,45 tok/s TG Prefill 1.174 · TTFT 2.313 ms | 1× | Radeon RX 7900 XTXAMD Ryzen Threadripper PRO 3955WX 16-Cores · 63 GB RAM | llama.cppgodclawQ4_K_M | Details → |