Across 11 shared benchmarks, Gemma 3 12B scores higher on 11 and Gemma 3 1B on 0. The widest gap is MMLU-Pro, where Gemma 3 12B scores 60.6 against 14.7.
| Benchmark | Gemma 3 12B | Gemma 3 1B |
|---|---|---|
| bbh | 85.7 | 39.1 |
| BIG-Bench Extra Hard | 16.3 | 7.2 |
| Global-MMLU-Lite | 69.5 | 34.2 |
| GPQA Diamond | 40.9 | 19.2 |
| GSM8K | 94.4 | 62.8 |
| humaneval | 85.4 | 41.5 |
| ifeval | 88.9 | 80.2 |
| LiveCodeBench | 24.6 | 1.9 |
| MBPP | 0.7 | 0.4 |
| MMLU-Pro | 60.6 | 14.7 |
| simpleqa | 6.3 | 2.2 |
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.