← Back to community benchmarks

gemma-4-26b-a4b-it

M3 Ultra (80c) · 512 GB · 4bit · 2026-04-17
Performance
16k
tokens
2,091
PP tok/s
59.0
TG tok/s
7836
TTFT (ms)
15.2
Peak mem (GB)
Hardware
Chip M3 Ultra (80c)
Memory 512 GB
GPU Cores 80
Software
oMLX v0.3.6
macOS macOS 15.3
Context 16,384
Performance by Context Length
Context PP tok/s TG tok/s Peak Mem
1k 1,271 96.6 14.2 GB view
4k 1,957 89.2 14.7 GB view
8k 2,134 75.9 14.8 GB view
16k 2,091 59.0 15.2 GB current
32k 1,879 45.4 16.0 GB view
64k 1,494 28.5 17.7 GB view
128k 1,022 15.9 21.2 GB view
195k 751.3 10.7 24.8 GB view
Batching Results
Batch Size TG tok/s Speedup
96.6 1.00×
137.7 1.43×
217.8 2.25×
268.7 2.78×