← Back to community benchmarks

DeepSeek-V3.2

M3 Ultra (80c) · 512 GB · 4bit · 2026-03-12
Performance
1k
tokens
209.8
PP tok/s
19.3
TG tok/s
4880
TTFT (ms)
354.2
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.2.9
macOS macOS 26.3.1
Context 1,024
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 209.8 19.3 354.2 GB current
4k 201.0 15.8 356.6 GB view
8k 169.7 15.1 360.1 GB view
16k 128.0 14.1 367.3 GB view
32k 84.0 12.8 382.2 GB view
Batching Results
Batch Size TG tok/s Speedup
19.3 1.00×
26.4 1.37×