← Back to community benchmarks

Qwen3.5-4B

M4 (10c) · 24 GB · 4bit · 2026-03-10
Performance
1k
tokens
402.5
PP tok/s
35.4
TG tok/s
2544
TTFT (ms)
3.9
Peak mem (GB)
Hardware
Chip M4 (10c)
Memory 24 GB
GPU Cores 10
Software
oMLX v0.2.6
macOS macOS 26.0.1
Context 1,024
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 402.5 35.4 3.9 GB current
4k 400.6 25.1 4.2 GB view
8k 387.4 30.7 4.6 GB view
16k 347.6 24.5 4.8 GB view
Batching Results
Batch Size TG tok/s Speedup
35.4 1.00×
68.7 1.94×
84.9 2.40×