Gemma 4 12B
57 records across 14 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| RTX PRO 6000 Blackwell Workstation Edition | Gemma 4 12B | Q4_K_M | llama.cpp | 287.33 | – | – | – | – | 2026-10-12 | sourced | – | source |
| RTX PRO 6000 Blackwell Workstation Edition | Gemma 4 12B | Q4_K_M | llama.cpp | 285.98 | – | – | – | – | 2026-10-12 | sourced | – | source |
| RTX 5090 | Gemma 4 12B | Q4_K_M | llama.cpp | 279.09 | – | – | – | – | 2026-10-12 | sourced | – | source |
| RTX 5090 | Gemma 4 12B | Q4_K_M | llama.cpp | 278.37 | – | – | – | – | 2026-10-12 | sourced | – | source |
| Arc B580 | Gemma 4 12B | QAT | llama.cpp (SYCL) | 100 | – | – | – | – | 2026-09 | community | – | source |
| GB10 | Gemma 4 12B | Q4_K_M | llama.cpp | 68.42 | – | – | – | – | 2026-10-12 | sourced | – | source |
| GB10 | Gemma 4 12B | Q4_K_M | llama.cpp | 67.31 | – | – | – | – | 2026-10-12 | sourced | – | source |
| RX 6800 | Gemma 4 12B | Q4_K_M | llama.cpp (Vulkan) | 52.1 | 799 | – | – | tg128 | 2026-10-01 | community | – | source |
| RX 6800 | Gemma 4 12B | Q4_K_M | llama.cpp (ROCm) | 48.9 | 736 | – | – | tg128 | 2026-10-01 | community | – | source |
| Arc B580 | Gemma 4 12B | QAT | llama.cpp (SYCL) | 40 | – | – | – | – | 2026-09 | community | – | source |
| Radeon RX 9060 XT | Gemma 4 12B | Q4_K_M | llama.cpp (Vulkan) | 39.0 | 1509 | – | – | tg128 | 2026-10-02 | community | – | source |
| M5 Max | Gemma 4 12B | 8bit | oMLX | 38.6 | 943.1 | – | – | 1024 | 2026-10-07 | community | – | source |
| M5 Max | Gemma 4 12B | 8bit | oMLX | 38.6 | 1181.0 | – | – | 1024 | 2026-10-09 | community | – | source |
| M5 Max | Gemma 4 12B | 8bit | oMLX | 38.5 | 1589.0 | – | – | 4096 | 2026-10-09 | community | – | source |
| M5 Max | Gemma 4 12B | 8bit | oMLX | 38.3 | 1505.0 | – | – | 4096 | 2026-10-11 | community | – | source |
| M5 Max | Gemma 4 12B | 8bit | oMLX | 38.2 | 907.3 | – | – | 1024 | 2026-10-11 | community | – | source |
| M5 Max | Gemma 4 12B | 8bit | oMLX | 38.0 | 1433.0 | – | – | 4096 | 2026-10-07 | community | – | source |
| Radeon RX 9060 XT | Gemma 4 12B | Q4_K_M | llama.cpp (ROCm) | 37.6 | 1467 | – | – | tg128 | 2026-10-02 | community | – | source |
| M5 Max | Gemma 4 12B | 8bit | oMLX | 37.6 | 1224.0 | – | – | 4096 | 2026-10-17 | community | – | source |
| M5 Max | Gemma 4 12B | 8bit | oMLX | 37.4 | 1610.0 | – | – | 16384 | 2026-10-17 | community | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 32.7 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 32.62 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 31.41 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 31.38 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 28.8 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 28.59 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 28.09 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 28.09 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 26.82 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 26.64 | – | – | – | – | 2026-10-11 | sourced | – | source |
| M4 Max | Gemma 4 12B | 8bit | oMLX | 25.1 | 412.4 | – | – | 16384 | 2026-10-11 | community | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 24.79 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 24.57 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 23.91 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 23.6 | – | – | – | – | 2026-10-11 | sourced | – | source |
| M5 Pro | Gemma 4 12B | nvfp4 | oMLX | 23.0 | 956.6 | – | – | 4096 | 2026-10-16 | community | – | source |
| M5 Pro | Gemma 4 12B | nvfp4 | oMLX | 23.0 | 907.9 | – | – | 16384 | 2026-10-16 | community | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 20.88 | – | – | – | – | 2026-10-11 | sourced | – | source |
| Radeon AI PRO R9700 | Gemma 4 12B | FP8 | vLLM | 20.65 | – | – | – | – | 2026-10-11 | sourced | – | source |
| M6 | Gemma 4 12B 12B | QAT | LM Studio | 19 | – | – | – | – | 2026-10 | community | – | source |
| M4 Pro | Gemma 4 12B | 8bit | oMLX | 17.3 | 274.5 | – | – | 1024 | 2026-10-07 | community | – | source |
| M4 Pro | Gemma 4 12B | 8bit | oMLX | 17.2 | 271.6 | – | – | 4096 | 2026-10-07 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | 13.93 | – | – | – | 4096 | 2026-10 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | 13.77 | – | – | – | 4096 | 2026-10 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | 13.33 | – | – | – | 16384 | 2026-10 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | 12.58 | – | – | – | 32768 | 2026-10 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | 12.15 | – | – | – | 16384 | 2026-10 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | 10.66 | – | – | – | 32768 | 2026-10 | community | contradiction | source |
| M3 Pro | Gemma 4 12B | 8bit | oMLX | 10.2 | 213.8 | – | – | 1024 | 2026-10-14 | community | – | source |
| M3 Pro | Gemma 4 12B | 8bit | oMLX | 10.2 | 212.7 | – | – | 1024 | 2026-10-14 | community | – | source |
| M3 Pro | Gemma 4 12B | 8bit | oMLX | 10.1 | 216.5 | – | – | 1024 | 2026-10-14 | community | – | source |
| M3 Pro | Gemma 4 12B | 8bit | oMLX | 10.1 | 210.9 | – | – | 1024 | 2026-10-14 | community | – | source |
| M3 Pro | Gemma 4 12B | 8bit | oMLX | 10.1 | 215.2 | – | – | 4096 | 2026-10-14 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | – | 261.5 | – | – | – | 2026-10 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | – | 263.2 | – | – | – | 2026-10 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | – | 186.7 | – | – | 16384 | 2026-10 | community | – | source |
| Adreno X2-90 | Gemma 4 12B | Q4_0 | llama.cpp (OpenCL) | – | 209.3 | – | – | 16384 | 2026-10 | community | – | source |