13 records across 4 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|
| M4 Ultra (192 GB) | Qwen 3 32B | Q4_K - Medium | Ollama | 45 | – | – | 1 | – | – | sourced | – | source |
| M4 Max (40-core GPU, 64 GB) | Qwen 3 32B | Q4_K - Medium | Ollama | 22 | – | – | – | 128000 | – | community | – | source |
| M4 Max (128 GB) | Qwen 3 32B | Q4_K - Medium | LM Studio | 11.72 | – | – | 0.81 | 10000 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 10.41 | 153.63 | – | 1.72 | 264 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 10.29 | 164.83 | – | 4.39 | 723 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 10.28 | 171.37 | – | 2.63 | 450 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 10.19 | 169.15 | – | 7.21 | 1219 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 10.09 | 166.81 | – | 11.14 | 1858 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 9.89 | 162.22 | – | 18.36 | 2979 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 9.67 | 154.16 | – | 30.29 | 4669 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 9.2 | 140.11 | – | 56.73 | 7948 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 8.6 | 127.96 | – | 97.03 | 12416 | – | sourced | – | source |
| M3 Max (GPU count not published, 64 GB) | Qwen 3 32B | Q8_0 | llama.cpp | 7.58 | 111.18 | – | 181.44 | 20172 | – | sourced | – | source |