← Back to community benchmarks

DeepSeek-V4-Flash-oQ8-mtp

M3 Ultra (80c) · 512 GB · 8bit · 2026-07-17
Performance
16k
tokens
482.8
PP tok/s
25.6
TG tok/s
33939
TTFT (ms)
146.0
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.5.1
macOS macOS 26.5.2
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 525.5 28.0 145.5 GB view
4k 509.1 26.8 145.5 GB view
8k 489.9 26.1 145.6 GB view
16k 482.8 25.6 146.0 GB current
32k 467.3 24.8 146.8 GB view
64k 440.5 23.9 148.3 GB view
128k 399.1 21.4 151.3 GB view