← Back to community benchmarks

Qwen3-4B-Instruct-2507

M4 Pro (16c) · 48 GB · 4bit · 2026-04-24
Performance
8k
tokens
542.3
PP tok/s
48.1
TG tok/s
15106
TTFT (ms)
3.8
Peak mem (GB)
Hardware
Chip M4 Pro (16c)
Memory 48 GB
GPU Cores 16
Software
oMLX v0.3.6
macOS macOS 26.5
Context 8,192
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 636.7 81.1 2.9 GB view
4k 605.1 63.7 3.4 GB view
8k 542.3 48.1 3.8 GB current
16k 437.9 31.9 5.0 GB view
Batching Results
Batch Size TG tok/s Speedup
81.1 1.00×
125.0 1.54×
139.1 1.72×