← Back to community benchmarks

Qwen2.5-Coder-3B-Instruct

M3 Ultra (80c) · 512 GB · 4bit · 2026-04-30

Performance

4k

tokens

3,350

PP tok/s

151.1

TG tok/s

1223

TTFT (ms)

2.6

Peak mem (GB)

Hardware

Chip M3 Ultra (80c)

Memory 512 GB

GPU Cores 80

Software

oMLX v0.3.6

macOS macOS 26.4.1

Context 4,096

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	2,604	172.7	2.4 GB	view
4k	3,350	151.1	2.6 GB	current

Batching Results

Batch Size	TG tok/s	Speedup
1×	172.7	1.00×
2×	310.0	1.80×
4×	474.3	2.75×