Apple M3 Ultra
28 records across 3 models (quants: Q4_K_M, Q8_0, UD-Q4_K_XL).
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Apple M3 Ultra | GLM-5.3-Flash | UD-Q4_K_XL | Metal | 40 | – | – | – | 256 | 2026-09 | community | – | source |
| Apple M3 Ultra | GLM-5.3-Flash | UD-Q4_K_XL | Metal | 10.6 | – | – | – | 8000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 365.5 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 346.8 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 297.3 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 295.9 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 213.6 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 213.3 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 185.0 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 184.7 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 145.9 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 145.4 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 95.5 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.6-35B-A3B | Q4_K_M | llama.cpp | 95.1 | – | – | – | tg32 | 2026-10 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 78 | – | – | – | 256 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 60 | – | – | – | 256 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 45 | – | – | – | 256 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 32.8 | – | – | – | 256 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 25.9 | – | – | – | 16000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 25.7 | – | – | – | 32000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 23.5 | – | – | – | 8000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 23 | – | – | – | 4000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 11.6 | – | – | – | 8000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 10.3 | – | – | – | 16000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 8.6 | – | – | – | 32000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 6.9 | – | – | – | 16000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 5.5 | – | – | – | 32000 | 2026-09 | community | – | source |
| Apple M3 Ultra | Qwen3.8-Flash-Next | Q8_0 | Metal | 5.2 | – | – | – | 16000 | 2026-09 | community | – | source |
All rows cite their source. Provenance: community 40018 sourced 1085