🏆
Rankings
LLM Leaderboard
Real measurements of every tested language model on real, named hardware – rated by raw speed (performance in tokens per second, prefill and time to first token) and by practical task quality in complete agent and chat runs (harness). Pick a benchmark type below or filter by model, maker and hardware to see exactly what is tested and how the results are produced.
⚡ Performance (tok/s)🤖 Harness quality👥 Concurrency🖥️ real hardware
| # | Model / Maker | Metrics | Parallel | GPU / CPU / RAM | Runtime | ||
|---|---|---|---|---|---|---|---|
| 1 | gemma-4-31B-it31BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 174,07 tok/s TG Prefill 1.206 · TTFT 16.925 ms | 10× | NVIDIA GeForce RTX 3090 TiAMD Ryzen 5 5600X 6-Core Processor · 30 GB RAM | llamacpp | Details → | |
| 2 | gemma-4-31B-it31BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 156,49 tok/s TG Prefill 182 · TTFT 97.092 ms | 10× | NVIDIA GeForce RTX 3090 TiAMD Ryzen 5 5600X 6-Core Processor · 30 GB RAM | llama.cpp q3_k_m, q4-kv | Details → | |
| 3 | gemma-4-31B-it31BQ4_K_MGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 101,94 tok/s TG Prefill 176 · TTFT 112.412 ms | 10× | 2x AMD Radeon PRO W7900 Dual SlotAMD Ryzen Threadripper PRO 5975WX 32-Cores · 247 GB RAM | llama.cppgodclaw | Details → | |
| 4 | gemma-4-31B-it31BQ8_0Google Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 99,95 tok/s TG Prefill 166 · TTFT 119.519 ms | 10× | 2x AMD Radeon PRO W7900 Dual SlotAMD Ryzen Threadripper PRO 5975WX 32-Cores · 247 GB RAM | llama.cppgodclaw | Details → | |
| 5 | gemma-4-31B-it31BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 81,35 tok/s TG Prefill 998 · TTFT 9.924 ms | 5× | NVIDIA GeForce RTX 3090 TiAMD Ryzen 5 5600X 6-Core Processor · 30 GB RAM | llamacpp | Details → | |
| 6 | gemma-4-31B-it31BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 74,79 tok/s TG Prefill 1.062 · TTFT 9.322 ms | 5× | NVIDIA GeForce RTX 3090 TiAMD Ryzen 5 5600X 6-Core Processor · 30 GB RAM | llama.cpp q3_k_m, q4-kv | Details → | |
| 7 | gemma-4-31B-it31BQ8_0Google Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 72,98 tok/s TG Prefill 351 · TTFT 36.215 ms | 5× | 2x AMD Radeon PRO W7900 Dual SlotAMD Ryzen Threadripper PRO 5975WX 32-Cores · 247 GB RAM | llama.cppgodclaw | Details → | |
| 8 | gemma-4-31B-it31BQ4_K_MGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 54,15 tok/s TG Prefill 435 · TTFT 38.812 ms | 5× | 2x AMD Radeon PRO W7900 Dual SlotAMD Ryzen Threadripper PRO 5975WX 32-Cores · 247 GB RAM | llama.cppgodclaw | Details → | |
| 9 | gemma-4-31B-it31BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 30,10 tok/s TG Prefill 1.020 · TTFT 1.964 ms | 1× | NVIDIA GeForce RTX 3090 TiAMD Ryzen 5 5600X 6-Core Processor · 30 GB RAM | llama.cpp q3_k_m, q4-kv | Details → | |
| 10 | gemma-4-31B-it31BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 28,74 tok/s TG Prefill 938 · TTFT 2.110 ms | 1× | NVIDIA GeForce RTX 3090 TiAMD Ryzen 5 5600X 6-Core Processor · 30 GB RAM | llamacpp | Details → | |
| 11 | gemma-4-31B-it31BQ4_K_MGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 24,81 tok/s TG Prefill 701 · TTFT 2.818 ms | 1× | 2x AMD Radeon PRO W7900 Dual SlotAMD Ryzen Threadripper PRO 5975WX 32-Cores · 247 GB RAM | llama.cppgodclaw | Details → | |
| 12 | gemma-4-31B-it31BINT8Google Performance benchmarkPerformancetest Small 1.0 | 24,36 tok/s TG Prefill 309 · TTFT 91 ms | 1× | 3x AMD Radeon AI PRO R9700AMD Ryzen Threadripper PRO 7955WX 16-Cores · 184 GB RAM | vLLMgodclaw | Details → | |
| 13 | gemma-4-31B-it31BINT8Google Performance benchmark | 24,36 tok/s TG Prefill 309 · TTFT 91 ms | 1× | 3x AMD Radeon AI PRO R970032x AMD Ryzen Threadripper PRO 7955WX 16-Cores · 184 GB RAM | godclaw | Details → | |
| 14 | gemma-4-31B-it31BGoogle Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 19,50 tok/s TG Prefill 339 · TTFT 80.766 ms | 5× | NVIDIA GeForce RTX 3090 TiAMD Ryzen 9 8945HX with Radeon Graphics · 92 GB RAM | llmcodex_cli | Details → | |
| 15 | gemma-4-31B-it31BQ8_0Google Performance benchmarkTimebench 3 - Kombi (Prefill + Generation) | 19,44 tok/s TG Prefill 276 · TTFT 6.828 ms | 1× | 2x AMD Radeon PRO W7900 Dual SlotAMD Ryzen Threadripper PRO 5975WX 32-Cores · 247 GB RAM | llama.cppgodclaw | Details → | |
| 16 | gemma-4-31B-it31BINT8Google Performance benchmarkPerformancetest Small 1.0 | 12,45 tok/s TG Prefill 208 · TTFT 135 ms | 1× | 3x AMD Radeon AI PRO R9700AMD Ryzen Threadripper PRO 7955WX 16-Cores · 184 GB RAM | vLLMgodclaw | Details → | |
| 17 | gemma-4-31B-it31BINT8Google Performance benchmark | 12,45 tok/s TG Prefill 208 · TTFT 135 ms | 1× | 3x AMD Radeon AI PRO R970032x AMD Ryzen Threadripper PRO 7955WX 16-Cores · 184 GB RAM | godclaw | Details → | |
| 18 | gemma-4-31B-it31BINT8Google Performance benchmarkTimebench 2 - Token-Generation / Output-Durchsatz | 11,19 tok/s TG Prefill 162 · TTFT 716 ms | 1× | 3x AMD Radeon AI PRO R9700AMD Ryzen Threadripper PRO 7955WX 16-Cores · 184 GB RAM | vLLMgodclaw | Details → | |
| 19 | gemma-4-31B-it31BINT8Google Performance benchmark | 11,19 tok/s TG Prefill 162 · TTFT 716 ms | 1× | 3x AMD Radeon AI PRO R970032x AMD Ryzen Threadripper PRO 7955WX 16-Cores · 184 GB RAM | godclaw | Details → | |
| 20 | gemma-4-31B-it31BINT8Google Performance benchmarkPerformancetest Small 1.0 | 10,54 tok/s TG Prefill 199 · TTFT 141 ms | 1× | 3x AMD Radeon AI PRO R9700AMD Ryzen Threadripper PRO 7955WX 16-Cores · 184 GB RAM | vLLMgodclaw | Details → | |
| 21 | gemma-4-31B-it31BINT8Google Performance benchmark | 10,54 tok/s TG Prefill 199 · TTFT 141 ms | 1× | 3x AMD Radeon AI PRO R970032x AMD Ryzen Threadripper PRO 7955WX 16-Cores · 184 GB RAM | godclaw | Details → | |
| 22 | gemma-4-31B-it31BGoogle Harness benchmarkTool Usage Standard 1.0 | 298 Pkt 10,0% · 16/16 Aufg. · 9.625,0 s | 1× | AMD GPUAMD RYZEN AI MAX+ 395 w/ Radeon 8060S · 123 GB RAM | openclaw_cli | Details → |