Model benchmarks
gemma-3-4b-it local LLM performance
As of October 2026, gemma-3-4b-it runs at up to 12.9 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).
Model size
4B
Peak speed
12.9 tok/s
Average speed
12.8 tok/s
Avg PP
547.4 tok/s
Min memory
12.2 GB
Max context
16,384 tokens
Avg output / run
6,317 tokens
Avg runtime / run
8m 34s
Avg quality
50.7
Benchmark runs
2
GPUs tested
1
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 2 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 | 47.2 | 50.7 | 54.3 |
| Agent Workflow | 32.4 | 47.2 | 62.1 |
| Code Generation | 32.1 | 32.9 | 33.6 |
| Role Play & Narrative | 74.7 | 75.9 | 77.1 |
| Research & Analysis | 45.9 | 47.0 | 48.1 |
Performance by hardware and tool
Every hardware/tool/quantization combination gemma-3-4b-it 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 |
|---|---|---|---|---|---|---|---|---|
| AMD Radeon Vega Series / Radeon Vega Mobile Series | Bionic | — | 12.9 tok/s | 12.8 tok/s | 12.2 GB | 16,384 tokens | 50.7 | 2 |
Benchmark runs
All 2 gemma-3-4b-it runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is gemma-3-4b-it good for coding?
- In our benchmarks, gemma-3-4b-it scores 32.9/100 for coding. It runs at about 12.8 tok/s, so if you want more speed, tensorfold-qwen38-27b is faster (~25.0 tok/s) and still scores well for coding (84.3/100).
- Is gemma-3-4b-it good for agentic (tool-using) tasks?
- In our benchmarks, gemma-3-4b-it scores 47.2/100 for agentic workflows. It runs at about 12.8 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 gemma-3-4b-it for local inference?
- Across 2 community benchmark runs, gemma-3-4b-it reaches up to 12.9 tok/s and averages 12.8 tok/s, with the fastest results on AMD Radeon Vega Series / Radeon Vega Mobile Series.
- How much memory does gemma-3-4b-it need?
- The leanest observed configuration used about 12.2 GB of memory.
- Which tools have been used to run gemma-3-4b-it?
- Benchmarks were submitted using Bionic. Results are community-contributed and updated as new runs arrive.