Is gemma4:31b good for coding?

In our benchmarks, gemma4:31b scores 63.1/100 for coding. It runs at about 5.1 tok/s, so if you want more speed, Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_XXS-mtp is faster (~37.5 tok/s) and still scores well for coding (83.8/100).

Is gemma4:31b good for agentic (tool-using) tasks?

In our benchmarks, gemma4:31b scores 82.4/100 for agentic workflows. It runs at about 5.1 tok/s, so if you want more speed, muse-glimmer:latest is faster (~33.3 tok/s) and still scores well for agentic workflows (91.6/100).

How fast is gemma4:31b for local inference?

Across 11 community benchmark runs, gemma4:31b reaches up to 7.5 tok/s and averages 5.1 tok/s, with the fastest results on Apple M5 Max.

How much memory does gemma4:31b need?

The leanest observed configuration used about 20.2 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.

Model benchmarks

gemma4:31b local LLM performance

As of September 2026, gemma4:31b runs at up to 7.5 tok/s for local inference (best of 11 community benchmark runs across 4 GPUs).

OllamaQ4_K_M

ShareReddit X

Model size

31.3B

Peak speed

7.5 tok/s

Average speed

5.1 tok/s

Avg PP

23.6 tok/s

Min memory

20.2 GB

Max context

65,536 tokens

Avg output / run

10,176 tokens

Avg runtime / run

42m 17s

Avg quality

73.9

Benchmark runs

GPUs tested

Quality by task

Average LLM-judged quality (0–100) with the run-to-run spread shown as a P5–P95 band, overall and for each benchmark task, across all 11 runs. The low and high columns show how much the judge’s score varies between runs, and need at least two runs to display.

Task	P5 (low)	Avg	P95 (high)
Overall	62.6	73.9	79.2
Agent Workflow	72.6	82.4	89.2
Code Generation	36.2	63.1	72.5
Role Play & Narrative	74.6	80.9	85.6
Research & Analysis	57.1	69.1	78.5

Performance by hardware and tool

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

Hardware	Tool	Quant	Best tok/s	Avg tok/s	Memory	Context	Quality	Runs
Apple M5 Max	Ollama	Q4_K_M	7.5 tok/s	7.4 tok/s	28.4 GB	65,536 tokens	77.2	2
Apple M4 Pro	Ollama	Q4_K_M	6.0 tok/s	6.0 tok/s	20.2 GB	8,192 tokens	76.6	1
AMD Radeon RX 7900 XTX	Ollama	Q4_K_M	5.5 tok/s	4.7 tok/s	20.6 GB	65,536 tokens	72.4	7
Apple M1 Pro	Ollama	Q4_K_M	2.4 tok/s	2.4 tok/s	20.8 GB	65,536 tokens	75.3	1

Benchmark runs

All 11 gemma4:31b runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Open run page →

Frequently asked questions

Is gemma4:31b good for coding?: In our benchmarks, gemma4:31b scores 63.1/100 for coding. It runs at about 5.1 tok/s, so if you want more speed, Qwen3.8-27B-Turbo-Brilliance-50X-RI-GSQ-RCO-IQ3_XXS-mtp is faster (~37.5 tok/s) and still scores well for coding (83.8/100).
Is gemma4:31b good for agentic (tool-using) tasks?: In our benchmarks, gemma4:31b scores 82.4/100 for agentic workflows. It runs at about 5.1 tok/s, so if you want more speed, muse-glimmer:latest is faster (~33.3 tok/s) and still scores well for agentic workflows (91.6/100).
How fast is gemma4:31b for local inference?: Across 11 community benchmark runs, gemma4:31b reaches up to 7.5 tok/s and averages 5.1 tok/s, with the fastest results on Apple M5 Max.
How much memory does gemma4:31b need?: The leanest observed configuration used about 20.2 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.