gpt-sw3-1.3B: benchmark results
AI Sweden's gpt-sw3-1.3b, 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 1173 ± 29, rank #9753 of 10916 rated models, from 43 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Faroese NLU - FoSent | 15.51 | Matthews correlation coefficient (%) | 37.8 |
| EuroEval Danish NLU - Multi Wiki QA DA | 56.18 | Character F1 (%) | 36.8 |
| EuroEval Swedish NLU - Multi Wiki QA SV | 52.74 | Character F1 (%) | 36.4 |
| EuroEval Norwegian Knowledge - Idioms NO | 0.47 | MCC (x100) | 29.1 |
| EuroEval Icelandic Common Sense Reasoning | 0.89 | Common Sense Reasoning Average Score (%) | 22 |
| EuroEval Norwegian Knowledge | 1.94 | Knowledge Average Score (%) | 16.8 |
| EuroEval Norwegian Knowledge - NRK Quiz QA | 3.42 | MCC (x100) | 16.6 |
| EuroEval Danish NLU - ScaLA DA | 0.31 | Matthews correlation coefficient (%) | 15.8 |
| EuroEval Icelandic Knowledge | 0.58 | Knowledge Average Score (%) | 15.3 |
| EuroEval Danish NLU | 16.51 | NLU Average Score (%) | 14.7 |
| EuroEval Swedish Knowledge | 1.08 | Knowledge Average Score (%) | 12.8 |
| Open LLM Leaderboard v1 - TruthfulQA MC2 | 39.97 | MC2 (%) (0-shot) | 12.6 |
Interactive version: theaggregate.ai/model?slug=gpt-sw3-1-3b · How It Works · Data refreshed daily, snapshot 2026-11-08.