gpt-sw3-126m: benchmark results
AI Sweden's gpt-sw3-126m, a decoder-only causal language model developed with RISE and WASP WARA for Media and Language using NeMo Megatron GPT. Its card describes Nordic-language, English and programming-code pretraining. Provider: AI Sweden. Access: Open.
Aggregate Rating 1196 ± 38, rank #9641 of 10916 rated models, from 91 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Icelandic NLU - ScaLA IS | 1.69 | Matthews correlation coefficient (%) | 41.8 |
| EuroEval Icelandic Common Sense Reasoning | 1.46 | Common Sense Reasoning Average Score (%) | 28 |
| EuroEval Faroese NLU - FoSent | 5.21 | Matthews correlation coefficient (%) | 25.6 |
| EuroEval Spanish NLU - ScaLA ES | 0.41 | Matthews correlation coefficient (%) | 23.4 |
| EuroEval Danish Knowledge - Danske Talemaader | 2.48 | MCC (x100) | 18.2 |
| EuroEval Norwegian Common Sense Reasoning | 1.33 | Common Sense Reasoning Average Score (%) | 17.6 |
| EuroEval Danish Common Sense Reasoning | 1.56 | Common Sense Reasoning Average Score (%) | 16.6 |
| EuroEval Dutch NLU - ScaLA NL | 1.74 | Matthews correlation coefficient (%) | 16 |
| EuroEval Faroese NLU - ScaLA FO | 0 | Matthews correlation coefficient (%) | 15.9 |
| EuroEval Norwegian Knowledge - Idioms NO | -0.16 | MCC (x100) | 15.8 |
| EuroEval Danish NLU - ScaLA DA | 0.26 | Matthews correlation coefficient (%) | 15.1 |
| EuroEval English Knowledge | 3.88 | Knowledge Average Score (%) | 14.2 |
Interactive version: theaggregate.ai/model?slug=gpt-sw3-126m · How It Works · Data refreshed daily, snapshot 2026-11-08.