Configs › Qwen3.8-27B › Qwen3.8-27B on Mac Max: 41.5 tok/s, 4 measured configs
Qwen3.8-27B on Mac Max: 41.5 tok/s, 4 measured configs
Measured Qwen3.8-27B speeds on Mac Max: fastest 41.5 tok/s single-stream (llama.cpp (speculative decode, mtp)), 4 configs with quant, backend and context, each linked to its source, snapshot 2026-10-29.
Snapshot 2026-10-29 · 413 configs on AI boxes and 288 on GPU cards · 121 models. Every number is generated from the tracker snapshot and linked to where it was measured; nothing here is typed by hand.
Fastest measured single-stream run of Qwen3.8-27B on Mac Max: 41.5 tok/s (llama.cpp (speculative decode, mtp)), snapshot 2026-10-29.
Every measured single-stream run of Qwen3.8-27B on Mac Max
| Decode tok/s | Quant | Backend | Context | Trust | Measured by | Source checked | Source |
|---|---|---|---|---|---|---|---|
| 41.5 | llama.cpp (speculative decode, mtp) | 16000 | Extracted, source-linked | xthomaswang | 2026-10-19 | source | |
| 31.5 | MLX-4BIT | mlx | Extracted, source-linked | lizhun781021 | 2026-09-29 | source | |
| 27.4 | mlx | 16000 | Extracted, source-linked | xthomaswang | 2026-10-19 | source | |
| 26.2 | vllm/metal | 16000 | Extracted, source-linked | xthomaswang | 2026-10-19 | source |
Backends on Mac Max
Fastest plain-decoding run per backend. Quants can differ between rows.
- mlx: 31.5 tok/s (MLX-4BIT)
- vllm/metal: 26.2 tok/s
Qwen3.8-27B on other boxes
- Mac Ultra: 42.6 tok/s best of 1 config
- Strix Halo: 36.8 tok/s best of 19 configs
- DGX Spark: 23.6 tok/s best of 2 configs
Questions and answers
- How fast is Qwen3.8-27B on Mac Max?
- The fastest of 4 measured configs is 41.5 tok/s single-stream (llama.cpp (speculative decode, mtp)). The slowest is 26.2 tok/s.
- Which backend is fastest for Qwen3.8-27B on Mac Max?
- mlx at 31.5 tok/s, then vllm/metal at 26.2. Each backend figure is its best plain-decoding row and may use a different quant.
- How does Mac Max compare with other boxes for Qwen3.8-27B?
- Mac Max best 41.5 tok/s. Other boxes, best measured: Mac Ultra 42.6, Strix Halo 36.8, DGX Spark 23.6. Backends and quants differ, so this is the best known number per box, not a controlled test.
See also: Qwen3.8-27B on every box · Mac Max hardware page (from S$3,499 in Singapore today).
Last updated: snapshot 2026-10-29; newest source check 2026-10-19.
More: all configs · hardware · news · contributors · methodology · llms.txt · llms-full.txt · API (OpenAPI) · JSON snapshot. Built by Altronis, private on-prem AI, Singapore.