← Back to community benchmarks

Qwen3.5-4B

M3 Max (30c) · 96 GB · 4bit · 2026-04-19
Performance
8k
tokens
988.3
PP tok/s
79.2
TG tok/s
8289
TTFT (ms)
4.5
Peak mem (GB)
Hardware
Chip M3 Max (30c)
Memory 96 GB
GPU Cores 30
Software
oMLX v0.3.6
macOS macOS 26.4.1
Context 8,192
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 874.5 91.4 3.4 GB view
4k 994.9 86.2 4.1 GB view
8k 988.3 79.2 4.5 GB current
16k 879.7 68.1 5.2 GB view
Batching Results
Batch Size TG tok/s Speedup
91.4 1.00×
156.8 1.72×
206.7 2.26×