← Back to community benchmarks

DeepSeek-V4-Flash

M3 Ultra (60c) · 256 GB · 8bit · 2026-05-20
Performance
32k
tokens
313.7
PP tok/s
25.3
TG tok/s
104448
TTFT (ms)
146.7
Peak mem (GB)
Hardware
Chip M3 Ultra (60c)
Memory 256 GB
GPU Cores 60
Software
oMLX v0.3.9rc1
macOS macOS 26.2
Context 32,768
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 356.0 29.6 145.6 GB view
4k 354.1 27.5 145.8 GB view
8k 343.6 27.1 145.7 GB view
16k 332.2 26.1 146.0 GB view
32k 313.7 25.3 146.7 GB current
Batching Results
Batch Size TG tok/s Speedup
29.6 1.00×
44.3 1.50×
63.3 2.14×