Gemma 3 12B PT
Gemma 3 12B PT places in the top three on 6 benchmarks.
27
Benchmarks with a score
0
Ranked first, out of five or more models
6
In the top three
Gemma 3 12B PT
Open weights Mid-size
No Epoch Capabilities Index score yet
- Angry Tweets 53% #76 /149
- FoQA 77.3% #5 /91
- GermanQuAD 68.4% #4 /88
How does it do against its rivals?
Gemma 3 12B PT beats Aya Expanse 32B on 7 of the 13 benchmarks they share. Apertus 70B does best against it, ahead on 5 of 8.
-
Aya Expanse 32B 7 ahead, 6 behind Compare
-
Qwen3-14B 9 ahead, 8 behind, 1 tied Compare
-
Voxtral Small 4 ahead, 4 behind Compare
-
Phi-4 6 ahead, 7 behind Compare
-
Qwen3-32B 5 ahead, 8 behind Compare
-
Apertus 70B 3 ahead, 5 behind Compare
Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| EuroEval Faroese Grammatical Error Correction | 12.9% | #3 of 71 | Official leaderboard | |
| GermanQuAD | 68.4% | #4 of 88 | Official leaderboard | |
| FoQA | 77.3% | #5 of 91 | Official leaderboard | |
| GuiaCat | 67.5% | #3 of 35 | Official leaderboard | |
| BLiMP-NL | 48.5% | #2 of 21 | Official leaderboard | |
| Dutch CoLA | 48.5% | #2 of 21 | Official leaderboard | |
| DaLA | 52.3% | #3 of 25 | Official leaderboard | |
| MultiBLiMP | 88.8% | #3 of 25 | Official leaderboard | |
| Open Hebrew LLM Leaderboard | 64.0 | #15 of 88 | Official leaderboard | |
| BLUEX | 68.2% | #38 of 177 | Official leaderboard | |
| GreekMMLU | 77.3% | #8 of 34 | Official leaderboard | |
| Dutch Book Reviews Dataset | 90.7% | #28 of 105 | Official leaderboard | |
| Uncheatable Eval | 6.86 | #10 of 33 | Official leaderboard | |
| FoSent | 55.4% | #50 of 147 | Official leaderboard | |
| HeQ | 71.5% | #30 of 88 | Official leaderboard | |
| QIMMA | 53.2% | #18 of 44 | Official leaderboard | |
| Open Portuguese LLM Leaderboard | 68.6 | #74 of 175 | Official leaderboard | |
| Angry Tweets | 53% | #76 of 149 | Official leaderboard | |
| 3LM | 73.3% | #13 of 25 | Official leaderboard | |
| Global PIQA | 64.2% | #31 of 59 | Official leaderboard | |
| EuroEval | 2.37 | #46 of 75 | Official leaderboard | |
| NOVA-63 | 47.9% | #34 of 52 | Official leaderboard | |
| EuroEval ERRNews | 24.0 | #61 of 76 | Official leaderboard | |
| OrangeSum | 29.3% | #28 of 34 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CaBBQ | 50.5% | #6 of 11 | Official leaderboard | |
| EsBBQ | 47.5% | #8 of 11 | Official leaderboard | |
| nocot-bench (NCRI) | 49.9 | #176 of 201 | Official leaderboard |
Who measured it?
Independent runs cover all 27 of its benchmarks.
- An independent run only 27
Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards