← Back to community benchmarks

LongShuGameDev-Qwen3.5-122B-REAP-Architect

M3 Max (40c) · 64 GB · 4bit · 2026-04-01
Performance
16k
tokens
1,394
PP tok/s
14.1
TG tok/s
11756
TTFT (ms)
46.6
Peak mem (GB)
Hardware
Chip M3 Max (40c)
Memory 64 GB
GPU Cores 40
Software
oMLX v0.3.0
macOS macOS 26.3.1
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 449.6 50.6 43.1 GB view
4k 499.1 46.6 44.6 GB view
8k 488.2 40.2 45.2 GB view
16k 1,394 14.1 46.6 GB current
32k 1,336 13.8 48.1 GB view
64k 1,185 13.2 49.9 GB view
128k 803.8 12.2 53.4 GB view
195k 683.1 11.8 57.5 GB view
Batching Results
Batch Size TG tok/s Speedup
50.6 1.00×
65.3 1.29×
74.2 1.47×
85.3 1.69×