← Back to community benchmarks

Kimi-K2.63.3bit

M3 Ultra (80c) · 512 GB · 4bit · 2026-04-25

Performance

4k

tokens

243.6

PP tok/s

20.5

TG tok/s

16815

TTFT (ms)

400.9

Peak mem (GB)

Hardware

Chip M3 Ultra (80c)

Memory 512 GB

GPU Cores 80

Software

oMLX v0.3.7rc1

macOS macOS 15.7.4

Context 4,096

Performance by Context Length

Context	PP tok/s	TG tok/s	Peak Mem
1k	213.1	21.2	399.3 GB	view
4k	243.6	20.5	400.9 GB	current

Batching Results

Batch Size	TG tok/s	Speedup
1×	21.2	1.00×
2×	30.1	1.42×
4×	40.9	1.93×