← Back to community benchmarks

OpenAI-gpt-oss-20B-INSTRUCT-Heretic-Uncensored-MXFP4-q8-hi

M4 (10c) · 24 GB · 4bit · 2026-04-27
Performance
32k
tokens
300.8
PP tok/s
18.0
TG tok/s
108926
TTFT (ms)
12.5
Peak mem (GB)
Hardware
Chip M4 (10c)
Memory 24 GB
GPU Cores 10
Software
oMLX v0.3.7
macOS macOS 15.7.5
Context 32,768
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 342.4 33.3 11.8 GB view
4k 382.9 31.6 11.9 GB view
8k 374.8 29.4 12.0 GB view
16k 347.6 23.8 12.1 GB view
32k 300.8 18.0 12.5 GB current
Batching Results
Batch Size TG tok/s Speedup
33.3 1.00×
46.3 1.39×
54.9 1.65×