← Back to community benchmarks

Qwen3.5-122B-A10B-oQ4+

M3 Ultra (80c) · 512 GB · 4bit · 2026-03-22

Performance

1k

tokens

637.0

PP tok/s

42.3

TG tok/s

1608

TTFT (ms)

62.5

Peak mem (GB)

Hardware

Chip M3 Ultra (80c)

Memory 512 GB

GPU Cores 80

Software

oMLX v0.2.20.dev3

macOS macOS 26.3.1

Context 1,024

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	637.0	42.3	62.5 GB	current
4k	760.4	39.8	64.0 GB	view

Batching Results

Batch Size	TG tok/s	Speedup
1×	42.3	1.00×
2×	69.8	1.65×
4×	104.1	2.46×