← Back to community benchmarks

DeepSeek-R1-0528-Qwen3-8B

M3 Max (30c) · 36 GB · 8bit · 2026-05-14

Performance

4k

tokens

556.0

PP tok/s

30.3

TG tok/s

7367

TTFT (ms)

9.4

Peak mem (GB)

Hardware

Chip M3 Max (30c)

Memory 36 GB

GPU Cores 30

Software

oMLX v0.3.9.dev2

macOS macOS 26.3.1

Context 4,096

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	546.4	32.5	8.9 GB	view
4k	556.0	30.3	9.4 GB	current

Batching Results

Batch Size	TG tok/s	Speedup
1×	32.5	1.00×
2×	60.7	1.87×
4×	82.1	2.53×