Benchmarks · community-measured
qwen3.8-27b-4bit on Apple M4 Max
Real measured local-inference performance for
qwen3.8-27b-4bit on an Apple M4 Max (128 GB unified memory),
from the rapid-mlx community benchmark corpus. Headline submission
2026-09-08, measured on rapid-mlx 0.13.4.
28.9
tok/s decode (mtp)
260
tok/s prefill
8.2 s
first token
18.8 GB
peak memory
A single community run so far — treat it as one data point, not a distribution. Reproduce it below and submit yours to firm it up.
All measurements for this pair
| RAM | Accel | Decode | Prefill | TTFT | Peak mem | Runs | rapid-mlx | Latest |
|---|---|---|---|---|---|---|---|---|
| 128 GB | mtp | 28.9 tok/s | 260 tok/s | 8.2 s | 18.8 GB | 1 | 0.13.4 | 2026-09-08 |
Standardized single-request benchmark: 5 measured rounds after 1 discarded warmup, greedy sampling, prefix cache disabled. Raw corpus: /api/benchmarks.
Reproduce this number
Install rapid-mlx, run the same suite on your Mac, and submit the result — it lands on the community leaderboard and updates this page.
curl -fsSL https://rapidmlx.com/install.sh | bash
rapid-mlx bench qwen3.8-27b-4bit --tier speed --submit