← Back to community benchmarks

Qwen3.6-35B-A3B-4bit-MTPLX-Optimized-Speed

M3 Ultra (80c) · 512 GB · 4bit · 2026-05-10

Performance

4k

tokens

2,510

PP tok/s

90.1

TG tok/s

1632

TTFT (ms)

20.0

Peak mem (GB)

Hardware

Chip M3 Ultra (80c)

Memory 512 GB

GPU Cores 80

Software

oMLX v0.3.9.dev1

macOS macOS 15.3

Context 4,096

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	1,629	90.6	19.2 GB	view
4k	2,510	90.1	20.0 GB	current

Batching Results

Batch Size	TG tok/s	Speedup
1×	90.6	1.00×
2×	159.1	1.76×
4×	257.1	2.84×