← Back to community benchmarks

DeepSeek-V3.25bit

M3 Ultra (80c) · 512 GB · 5bit · 2026-03-14

Performance

4k

tokens

197.2

PP tok/s

15.1

TG tok/s

20772

TTFT (ms)

435.8

Peak mem (GB)

Hardware

Chip M3 Ultra (80c)

Memory 512 GB

GPU Cores 80

Software

oMLX v0.2.12b2

macOS macOS 15.5

Context 4,096

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	192.6	16.3	432.4 GB	view
4k	197.2	15.1	435.8 GB	current
8k	173.1	14.6	440.1 GB	view