← Back to community benchmarks

Qwen3.5-27B-Heretic

M5 (10c) · 32 GB · 6bit · 2026-04-12

Performance

4k

tokens

202.6

PP tok/s

5.1

TG tok/s

20215

TTFT (ms)

25.0

Peak mem (GB)

Hardware

Chip M5 (10c)

Memory 32 GB

GPU Cores 10

Software

oMLX v0.3.5.dev1

macOS macOS 26.3.1

Context 4,096

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	192.5	5.2	23.5 GB	view
4k	202.6	5.1	25.0 GB	current

Batching Results

Batch Size	TG tok/s	Speedup
1×	5.2	1.00×
2×	10.7	2.06×
4×	14.3	2.75×