← Back to community benchmarks

Qwen3-4B-Thinking-2507

M1 Pro (16c) · 32 GB · 8bit · 2026-03-11
Performance
8k
tokens
235.5
PP tok/s
22.3
TG tok/s
34793
TTFT (ms)
5.7
Peak mem (GB)
Hardware
Chip M1 Pro (16c)
Memory 32 GB
GPU Cores 16
Software
oMLX v0.2.7
macOS macOS 26.3.1
Context 8,192
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 267.8 31.9 4.7 GB view
4k 255.5 27.2 5.3 GB view
8k 235.5 22.3 5.7 GB current
Batching Results
Batch Size TG tok/s Speedup
31.9 1.00×
56.3 1.76×
63.0 1.97×
56.5 1.77×