Gemma4 26B
22 records across 4 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTX Pro 6000 Blackwell | Gemma4 26B 26B | Q4_K | llama.cpp | 196.9 | 9437.2 | – | – | 4096 | – | sourced | – | source |
| RTX 5090 | Gemma4 26B 26B | Q4_K | llama.cpp | 180.3 | 8799.2 | – | – | 4096 | – | sourced | – | source |
| RTX Pro 6000 Blackwell | Gemma4 26B 26B | Q4_K | llama.cpp | 172.7 | 8453.8 | – | – | 16384 | – | sourced | – | source |
| RTX Pro 6000 Blackwell | Gemma4 26B 26B | Q4_K | llama.cpp | 170.3 | 7107.6 | – | – | 32768 | – | sourced | – | source |
| RTX 5090 | Gemma4 26B 26B | Q4_K | llama.cpp | 167.2 | 7733.2 | – | – | 16384 | – | sourced | – | source |
| RTX Pro 6000 Blackwell | Gemma4 26B 26B | Q4_K | llama.cpp | 161.0 | 5379.6 | – | – | 65536 | – | sourced | – | source |
| RTX 5090 | Gemma4 26B 26B | Q4_K | llama.cpp | 159.4 | 6292.6 | – | – | 32768 | – | sourced | – | source |
| RTX 5090 | Gemma4 26B 26B | Q4_K | llama.cpp | 149.4 | 4360.6 | – | – | 65536 | – | sourced | – | source |
| RTX Pro 6000 Blackwell | Gemma4 26B 26B | Q4_K | llama.cpp | 133.2 | 3667.8 | – | – | 131072 | – | sourced | – | source |
| RTX 5090 | Gemma4 26B 26B | Q4_K | llama.cpp | 130.2 | 2839.1 | – | – | 131072 | – | sourced | – | source |
| RTX 3090 | Gemma4 26B 26B | Q4_K | llama.cpp | 119.4 | 3625.6 | – | – | 4096 | – | sourced | – | source |
| RTX 3090 | Gemma4 26B 26B | Q4_K | llama.cpp | 115.0 | 3068.9 | – | – | 16384 | – | sourced | – | source |
| RTX Pro 6000 Blackwell | Gemma4 26B 26B | Q4_K | llama.cpp | 112.5 | 2245.7 | – | – | 262144 | – | sourced | – | source |
| RTX 3090 | Gemma4 26B 26B | Q4_K | llama.cpp | 107.5 | 2453.4 | – | – | 32768 | – | sourced | – | source |
| RTX 5090 | Gemma4 26B 26B | Q4_K | llama.cpp | 106.0 | 1707.2 | – | – | 262144 | – | sourced | – | source |
| RTX 3090 | Gemma4 26B 26B | Q4_K | llama.cpp | 98.9 | 1765.1 | – | – | 65536 | – | sourced | – | source |
| RTX 3090 | Gemma4 26B 26B | Q4_K | llama.cpp | 83.0 | 1147.1 | – | – | 131072 | – | sourced | – | source |
| RTX 3090 | Gemma4 26B 26B | Q4_K | llama.cpp | 64.4 | 671.4 | – | – | 262144 | – | sourced | – | source |
| B70 Arc Pro | Gemma4 26B | Q4_K_M | Vulkan | – | 1525.36 | – | – | 8192 | 2026-10-02 | community | – | source |
| B70 Arc Pro | Gemma4 26B | Q4_K_M | Vulkan | – | 874.89 | – | – | 8192 | 2026-10-02 | community | – | source |
| B70 Arc Pro | Gemma4 26B | Q4_K_M | Vulkan | – | 1544.14 | – | – | 8192 | 2026-10-02 | community | – | source |
| B70 Arc Pro | Gemma4 26B | Q4_K_M | Vulkan | – | 880.44 | – | – | 8192 | 2026-10-02 | community | – | source |