← Back to community benchmarks

Qwen3.5-4B

M5 (10c) · 24 GB · 4bit · 2026-03-15
Performance
64k
tokens
670.5
PP tok/s
19.3
TG tok/s
97736
TTFT (ms)
8.0
Peak mem (GB)
Hardware
Chip M5 (10c)
Memory 24 GB
GPU Cores 10
Software
oMLX v0.2.9
macOS macOS 26.3.1
Context 65,536
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,211 30.8 3.9 GB view
4k 1,272 40.1 4.2 GB view
8k 1,278 40.9 4.6 GB view
16k 1,140 32.6 4.8 GB view
32k 938.3 27.7 5.9 GB view
64k 670.5 19.3 8.0 GB current
Batching Results
Batch Size TG tok/s Speedup
30.8 1.00×
95.1 3.09×
130.6 4.24×
134.9 4.38×