Benchmarks · community-measured

qwen3.5-9b-4bit on Apple M2 Pro

Real measured local-inference performance for qwen3.5-9b-4bit on an Apple M2 Pro (32 GB unified memory), from the rapid-mlx community benchmark corpus. Latest submission 2026-08-02, measured on rapid-mlx 0.11.9.

34.6
tok/s decode
170
tok/s prefill
12.5 s
first token
6.3 GB
peak memory

Aggregated from 3 community runs.

All measurements for this pair

RAMAccelDecodePrefillTTFTPeak memRunsrapid-mlxLatest
32 GBmtp35.1 tok/s172 tok/s12.4 s6.3 GB10.11.92026-08-02
32 GB34.6 tok/s170 tok/s12.5 s6.3 GB20.11.92026-08-02

Standardized single-request benchmark: 5 measured rounds after 1 discarded warmup, greedy sampling, prefix cache disabled. Raw corpus: /api/benchmarks.

Speculative decoding on this machine

A same-machine A/B pair measured mtp at 35.1 tok/s vs 34.9 tok/s baseline (speedup ×1.005).

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 qwen3.5-9b-4bit --tier speed --submit

More

qwen3.5-9b-4bit model page · Qwen family · Interactive leaderboard · Alias catalog