← Back to community benchmarks

Qwen3-Coder-Next-REAP-48B-A3B

M5 Pro (20c) · 64 GB · 4bit · 2026-05-04
Performance
128k
tokens
2,352
PP tok/s
53.9
TG tok/s
55730
TTFT (ms)
33.0
Peak mem (GB)
Hardware
Chip M5 Pro (20c)
Memory 64 GB
GPU Cores 20
Software
oMLX v0.3.8
macOS macOS 26.4.1
Context 131,072
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,426 78.6 26.6 GB view
4k 1,862 76.6 27.5 GB view
8k 1,870 73.7 28.1 GB view
16k 3,873 73.0 29.2 GB view
32k 3,673 70.1 29.9 GB view
64k 3,185 64.4 30.9 GB view
128k 2,352 53.9 33.0 GB current
Batching Results
Batch Size TG tok/s Speedup
78.6 1.00×
114.5 1.46×
157.3 2.00×
315.5 4.01×