Is gemma4:12b good for coding?

In our benchmarks, gemma4:12b scores 59.5/100 for coding. It runs at about 42.9 tok/s, so if you want more speed, nvidia/Qwen3.8-27B-NVFP4 is faster (~63.3 tok/s) and still scores well for coding (82.7/100).

Is gemma4:12b good for agentic (tool-using) tasks?

In our benchmarks, gemma4:12b scores 74.5/100 for agentic workflows. It runs at about 42.9 tok/s, so if you want more speed, Nex-N2.5-mini-APEX-Mini is faster (~88.2 tok/s) and still scores well for agentic workflows (91.3/100).

How fast is gemma4:12b for local inference?

Across 10 community benchmark runs, gemma4:12b reaches up to 108.4 tok/s and averages 42.9 tok/s, with the fastest results on NVIDIA GeForce RTX 4070 Ti.

How much memory does gemma4:12b need?

The leanest observed configuration used about 1.2 GB of memory (quantizations tested: Q4_K_M).

Which tools have been used to run gemma4:12b?

Benchmarks were submitted using Ollama. Results are community-contributed and updated as new runs arrive.

Model benchmarks

gemma4:12b local LLM performance

As of October 2026, gemma4:12b runs at up to 108.4 tok/s for local inference (best of 10 community benchmark runs across 8 GPUs).

OllamaQ4_K_M

ShareReddit X

Model size

11.9B

Peak speed

108.4 tok/s

Average speed

42.9 tok/s

Avg PP

848.0 tok/s

Min memory

1.2 GB

Max context

196,608 tokens

Avg output / run

14,606 tokens

Avg runtime / run

17m 9s

Avg quality

70.2

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 10 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	66.6	70.2	72.4
Agent Workflow	63.7	74.5	80.8
Code Generation	53.4	59.5	64.7
Role Play & Narrative	72.3	79.8	85.0
Research & Analysis	52.5	66.8	77.2

Performance by hardware and tool

Every hardware/tool/quantization combination gemma4:12b has been benchmarked on, ranked by peak token generation speed. Last updated October 2026.

Hardware	Tool	Quant	Best tok/s	Avg tok/s	Memory	Context	Quality	Runs
NVIDIA GeForce RTX 4070 Ti	Ollama	Q4_K_M	108.4 tok/s	108.4 tok/s	8.9 GB	65,536 tokens	71.1	1
AMD Radeon RX 7900 XTX	Ollama	Q4_K_M	63.1 tok/s	63.1 tok/s	8.2 GB	196,608 tokens	72.4	1
AMD Radeon RX 9060 XT	Ollama	Q4_K_M	51.0 tok/s	51.0 tok/s	1.2 GB	65,536 tokens	70.6	1
NVIDIA GeForce RTX 5060 Ti	Ollama	Q4_K_M	46.8 tok/s	46.8 tok/s	7.9 GB	65,536 tokens	67.2	1
CPU only	Ollama	Q4_K_M	46.3 tok/s	46.3 tok/s	7.8 GB	32,768 tokens	69.1	2
AMD Radeon RX 6900 XT	Ollama	Q4_K_M	34.2 tok/s	34.2 tok/s	7.8 GB	8,192 tokens	72.3	1
NVIDIA GeForce RTX 3060 Ti	Ollama	Q4_K_M	15.7 tok/s	15.7 tok/s	7.6 GB	8,192 tokens	66.1	1
Apple M1 Pro	Ollama	Q4_K_M	13.9 tok/s	13.9 tok/s	8.6 GB	65,536 tokens	71.1	1
NVIDIA GeForce GTX 1050 Ti	Ollama	Q4_K_M	3.2 tok/s	3.2 tok/s	15.6 GB	8,192 tokens	72.4	1

Benchmark runs

All 10 gemma4:12b runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.

Open run page →

Frequently asked questions

Is gemma4:12b good for coding?: In our benchmarks, gemma4:12b scores 59.5/100 for coding. It runs at about 42.9 tok/s, so if you want more speed, nvidia/Qwen3.8-27B-NVFP4 is faster (~63.3 tok/s) and still scores well for coding (82.7/100).
Is gemma4:12b good for agentic (tool-using) tasks?: In our benchmarks, gemma4:12b scores 74.5/100 for agentic workflows. It runs at about 42.9 tok/s, so if you want more speed, Nex-N2.5-mini-APEX-Mini is faster (~88.2 tok/s) and still scores well for agentic workflows (91.3/100).
How fast is gemma4:12b for local inference?: Across 10 community benchmark runs, gemma4:12b reaches up to 108.4 tok/s and averages 42.9 tok/s, with the fastest results on NVIDIA GeForce RTX 4070 Ti.
How much memory does gemma4:12b need?: The leanest observed configuration used about 1.2 GB of memory (quantizations tested: Q4_K_M).
Which tools have been used to run gemma4:12b?: Benchmarks were submitted using Ollama. Results are community-contributed and updated as new runs arrive.