← Back to community benchmarks

Qwen3.5-9B

M3 Ultra (80c) · 512 GB · 4bit · 2026-03-09
Performance
32k
tokens
1,315
PP tok/s
50.1
TG tok/s
24926
TTFT (ms)
8.7
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.2.6
macOS macOS 26.2
Context 32,768
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,465 105.7 6.7 GB view
4k 1,485 101.0 7.0 GB view
8k 1,465 96.2 7.1 GB view
16k 1,413 75.6 7.6 GB view
32k 1,315 50.1 8.7 GB current
64k 1,139 29.4 10.8 GB view
Batching Results
Batch Size TG tok/s Speedup
105.7 1.00×
181.9 1.72×
246.1 2.33×