gpt-sw3-6.7B: benchmark results
AI Sweden's gpt-sw3-6.7b, 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 1135 ± 36, rank #1611 of 1635 rated models, from 21 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Swedish NLU - Multi Wiki QA SV | 63.13 | Character F1 (%) | 43.1 |
| EuroEval Danish NLU - Multi Wiki QA DA | 67.2 | Character F1 (%) | 42.8 |
| EuroEval Norwegian Knowledge - Idioms NO | 0.26 | MCC (x100) | 24 |
| EuroEval Danish Knowledge | 13.1 | Knowledge Average Score (%) | 23.6 |
| Open Portuguese LLM - BLUEX | 25.73 | Accuracy (%) | 21.7 |
| Open Portuguese LLM - ASSIN2 RTE | 61.93 | Macro F1 (%) | 21.5 |
| EuroEval Danish Knowledge - Danish Citizen Tests | 13.1 | MCC (x100) | 20.8 |
| EuroEval Norwegian Knowledge - NRK Quiz QA | 5.39 | MCC (x100) | 20.8 |
| EuroEval Norwegian Knowledge | 2.82 | Knowledge Average Score (%) | 20.3 |
| Open Portuguese LLM - FaQuAD NLI | 43.97 | Macro F1 (%) | 17.1 |
| Open Portuguese LLM - ENEM | 21.13 | Accuracy (%) | 16.5 |
| Open LLM Leaderboard v1 - HellaSwag | 60.75 | Normalized accuracy (%) (10-shot) | 14.7 |
Interactive version: theaggregate.ai/model?slug=gpt-sw3-6-7b · How It Works · Data refreshed daily, snapshot 2026-11-05.