← Back to community benchmarks

Qwen3.5-4B

M5 Pro (16c) · 24 GB · 4bit · 2026-07-16

Performance

64k

tokens

583.5

PP tok/s

49.2

TG tok/s

112322

TTFT (ms)

6.0

Peak mem (GB)

Hardware

Chip M5 Pro (16c)

Memory 24 GB

GPU Cores 16

Software

oMLX v0.5.0

macOS macOS 26.5

Context 65,536

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
8k	996.4	81.4	4.0 GB	view
16k	970.6	74.8	4.5 GB	view
32k	799.2	62.3	4.9 GB	view
64k	583.5	49.2	6.0 GB	current
128k	350.4	31.6	8.0 GB	view