← Back to community benchmarks

Qwen3.5-2B-Distilled-OPUS-HereticVLM

M4 (10c) · 16 GB · 8bit · 2026-04-16

Performance

8k

tokens

997.9

PP tok/s

44.6

TG tok/s

8209

TTFT (ms)

3.2

Peak mem (GB)

Hardware

Chip M4 (10c)

Memory 16 GB

GPU Cores 10

Software

oMLX v0.3.5

macOS macOS 26.4.1

Context 8,192

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	876.3	45.5	2.6 GB	view
4k	994.2	45.3	3.0 GB	view
8k	997.9	44.6	3.2 GB	current

Batching Results

Batch Size	TG tok/s	Speedup
1×	45.5	1.00×
2×	93.9	2.06×
4×	186.7	4.10×