gpt-sw3-20B: benchmark results
AI Sweden's gpt-sw3-20b, 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. Released 2022-12-14. Access: Open.
Aggregate Rating 1190 ± 35, rank #1565 of 1635 rated models, from 49 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Danish NLU - Multi Wiki QA DA | 72.47 | Character F1 (%) | 57.6 |
| EuroEval Swedish NLU - Multi Wiki QA SV | 67.19 | Character F1 (%) | 48.2 |
| EuroEval Norwegian Knowledge - Idioms NO | 2.73 | MCC (x100) | 45.6 |
| EuroEval Swedish NLU - ScaLA SV | 12.95 | Matthews correlation coefficient (%) | 33.7 |
| EuroEval Danish NLU - ScaLA DA | 8.38 | Matthews correlation coefficient (%) | 31.8 |
| EuroEval Faroese NLU - FoSent | 8.4 | Matthews correlation coefficient (%) | 29.8 |
| EuroEval Norwegian Knowledge | 4.61 | Knowledge Average Score (%) | 27.5 |
| EuroEval Danish NLU | 26.95 | NLU Average Score (%) | 27 |
| EuroEval Danish Knowledge - Danish Citizen Tests | 19.87 | MCC (x100) | 25.8 |
| EuroEval Faroese NLU - ScaLA FO | 0.44 | Matthews correlation coefficient (%) | 24.4 |
| EuroEval Icelandic Knowledge | 1.67 | Knowledge Average Score (%) | 24.2 |
| Open Portuguese LLM - OAB Exams | 27.79 | Accuracy (%) | 22.8 |
Interactive version: theaggregate.ai/model?slug=gpt-sw3-20b · How It Works · Data refreshed daily, snapshot 2026-11-05.