Swift 1.5
15 records across 4 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| M5 Pro | Swift 1.5 | Q4/Q6/Q8 mixed | llama.cpp (Metal) | 67.52 | – | – | – | 262144 | 2026-10-02 | community | – | source |
| Radeon AI PRO R9700 | Swift 1.5 | MXFP4 | llama.cpp (ROCm) | 60.7 | – | – | – | – | 2026-10-03 | community | – | source |
| Radeon AI PRO R9700 | Swift 1.5 | MXFP4 | llama.cpp (ROCm) | 58.1 | – | – | – | – | 2026-10-03 | community | – | source |
| Radeon AI PRO R9700 | Swift 1.5 | MXFP4 | llama.cpp (ROCm) | 55.9 | – | – | – | – | 2026-10-03 | community | – | source |
| M5 Pro | Swift 1.5 | Q4/Q6/Q8 mixed | llama.cpp (Metal) | 34.42 | – | – | – | 262144 | 2026-10-02 | community | – | source |
| Radeon AI PRO R9700 | Swift 1.5 | MXFP4 | llama.cpp (ROCm) | 33.8 | 1206 | – | – | – | 2026-10-03 | community | – | source |
| Radeon AI PRO R9700 | Swift 1.5 | MXFP4 | llama.cpp (ROCm) | 33.3 | 1215 | – | – | – | 2026-10-03 | community | – | source |
| Radeon AI PRO R9700 | Swift 1.5 | MXFP4 | llama.cpp (ROCm) | 32.8 | 1212 | – | – | – | 2026-10-03 | community | – | source |
| M5 Pro | Swift 1.5 | Q4/Q6/Q8 mixed | llama.cpp (Metal) | 32.58 | – | – | – | 262144 | 2026-10-02 | community | – | source |
| M5 Pro | Swift 1.5 | Q4/Q6/Q8 mixed | llama.cpp (Metal) | 32.43 | – | – | – | 262144 | 2026-10-02 | community | – | source |
| M1 Ultra | Swift 1.5 | 4bit | oMLX | 30.7 | 208.5 | – | – | 1024 | 2026-10-15 | community | – | source |
| M1 Ultra | Swift 1.5 | 4bit | oMLX | 30.0 | 198.2 | – | – | 4096 | 2026-10-15 | community | – | source |
| M4 Pro | Swift 1.5 | 4bit | oMLX | 15.2 | 104.3 | – | – | 1024 | 2026-10-16 | community | – | source |
| M4 Pro | Swift 1.5 | 4bit | oMLX | 14.5 | 107.8 | – | – | 4096 | 2026-10-16 | community | – | source |
| M4 Pro | Swift 1.5 | 4bit | oMLX | 14.3 | 106.6 | – | – | 8192 | 2026-10-16 | community | – | source |