← Back to community benchmarks

Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated-hermes-oQ4-fp16-mtp

M2 Ultra (60c) · 64 GB · 4bit · 2026-07-23
Performance
128k
tokens
671.3
PP tok/s
49.8
TG tok/s
195261
TTFT (ms)
31.2
Peak mem (GB)
Hardware
Chip M2 Ultra (60c)
Memory 64 GB
GPU Cores 60
Software
oMLX v0.5.1
macOS macOS 26.5.2
Context 131,072
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,975 86.0 19.9 GB view
4k 2,060 83.5 20.7 GB view
8k 1,992 81.2 21.1 GB view
16k 1,837 77.3 21.7 GB view
32k 1,581 74.2 23.0 GB view
64k 1,153 63.4 25.7 GB view
128k 671.3 49.8 31.2 GB current
Batching Results
Batch Size TG tok/s Speedup
86.0 1.00×
143.4 1.67×
363.4 4.23×