Model benchmarks
google/gemma-3-4b local LLM performance
As of October 2026, google/gemma-3-4b runs at up to 24.6 tok/s for local inference (best of 2 community benchmark runs across 1 GPU).
Model size
4B
Peak speed
24.6 tok/s
Average speed
24.4 tok/s
Avg PP
748.5 tok/s
Min memory
4.7 GB
Max context
16,384 tokens
Avg output / run
6,441 tokens
Avg runtime / run
4m 38s
Avg quality
44.2
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 | 42.2 | 44.2 | 46.1 |
| Agent Workflow | 38.7 | 46.3 | 53.9 |
| Code Generation | 14.7 | 19.1 | 23.4 |
| Role Play & Narrative | 69.9 | 70.0 | 70.1 |
| Research & Analysis | 36.7 | 41.3 | 45.8 |
Performance by hardware and tool
Every hardware/tool/quantization combination google/gemma-3-4b 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 2050 | LM Studio | Q3_K_L | 24.6 tok/s | 24.4 tok/s | 4.7 GB | 16,384 tokens | 44.2 | 2 |
Benchmark runs
All 2 google/gemma-3-4b runs submitted so far — expand one for its hardware, quality breakdown and per-scenario detail.
Frequently asked questions
- Is google/gemma-3-4b good for coding?
- In our benchmarks, google/gemma-3-4b scores 19.1/100 for coding. It runs at about 24.4 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 google/gemma-3-4b good for agentic (tool-using) tasks?
- In our benchmarks, google/gemma-3-4b scores 46.3/100 for agentic workflows. It runs at about 24.4 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 google/gemma-3-4b for local inference?
- Across 2 community benchmark runs, google/gemma-3-4b reaches up to 24.6 tok/s and averages 24.4 tok/s, with the fastest results on NVIDIA GeForce RTX 2050.
- How much memory does google/gemma-3-4b need?
- The leanest observed configuration used about 4.7 GB of memory (quantizations tested: Q3_K_L).
- Which tools have been used to run google/gemma-3-4b?
- Benchmarks were submitted using LM Studio. Results are community-contributed and updated as new runs arrive.