Model benchmarks

gemma4:31b local LLM performance

As of May 2026, gemma4:31b runs at up to 7.5 tok/s for local inference (best of 10 community benchmark runs across 2 GPUs).

OllamaQ4_K_M

Model size

31.3B

Peak speed

7.5 tok/s

Average speed

5.3 tok/s

Min memory

23.2 GB

Max context

5,409 tokens

Best quality

92.8

Benchmark runs

10

GPUs tested

2

Performance by hardware and tool

Every hardware/tool/quantization combination gemma4:31b has been benchmarked on, ranked by peak token generation speed. Last updated May 2026.

HardwareToolQuantBest tok/sAvg tok/sMemoryContextQualityRuns
Apple M5 MaxOllamaQ4_K_M7.5 tok/s7.4 tok/s28.4 GB5,073 tokens76.02
AMD Radeon RX 7900 XTXOllamaQ4_K_M7.1 tok/s4.8 tok/s23.2 GB5,409 tokens66.28

Frequently asked questions

How fast is gemma4:31b for local inference?
Across 10 community benchmark runs, gemma4:31b reaches up to 7.5 tok/s and averages 5.3 tok/s, with the fastest results on Apple M5 Max.
How much memory does gemma4:31b need?
The leanest observed configuration used about 23.2 GB of memory (quantizations tested: Q4_K_M).
Which tools have been used to run gemma4:31b?
Benchmarks were submitted using Ollama. Results are community-contributed and updated as new runs arrive.