← Back to community benchmarks

Qwen3.5-35B-A3B-Hereticmxfp8

M1 Ultra (64c) · 64 GB · 8bit · 2026-03-12

Performance

4k

tokens

1,030

PP tok/s

52.7

TG tok/s

3975

TTFT (ms)

35.0

Peak mem (GB)

Hardware

Chip M1 Ultra (64c)

Memory 64 GB

GPU Cores 64

Software

oMLX v0.2.9

macOS macOS 26.3.1

Context 4,096

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	747.8	53.7	34.9 GB	view
4k	1,030	52.7	35.0 GB	current

Batching Results

Batch Size	TG tok/s	Speedup
1×	53.7	1.00×
2×	91.0	1.69×
4×	129.0	2.40×