Across 11 shared benchmarks, Gemma 3 1B scores higher on 0 and Gemma 3 4B on 11. The widest gap is LiveCodeBench, where Gemma 3 4B scores 12.6 against 1.9.
| Benchmark | Gemma 3 1B | Gemma 3 4B |
|---|---|---|
| bbh | 39.1 | 72.2 |
| BIG-Bench Extra Hard | 7.2 | 11 |
| Global-MMLU-Lite | 34.2 | 54.5 |
| GPQA Diamond | 19.2 | 30.8 |
| GSM8K | 62.8 | 89.9 |
| humaneval | 41.5 | 71.3 |
| ifeval | 80.2 | 90.2 |
| LiveCodeBench | 1.9 | 12.6 |
| MBPP | 0.4 | 0.6 |
| MMLU-Pro | 14.7 | 43.6 |
| simpleqa | 2.2 | 4 |
Best tracked score per model per benchmark (default configuration; source-attributed). ↓ marks lower-is-better metrics. Open either model for its full surface, provenance and pricing.