← Back to community benchmarks

gemma-4-31b-it

M3 Ultra (80c) · 512 GB · bf16 · 2026-04-03
Performance
16k
tokens
318.8
PP tok/s
8.3
TG tok/s
51391
TTFT (ms)
62.6
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.3.2
macOS macOS 26.4
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 327.7 9.6 60.0 GB view
4k 331.4 9.3 61.6 GB view
8k 329.7 9.0 61.7 GB view
16k 318.8 8.3 62.6 GB current
32k 295.8 7.3 65.0 GB view
64k 254.0 5.8 69.7 GB view
Batching Results
Batch Size TG tok/s Speedup
9.6 1.00×
11.0 1.15×
20.9 2.18×