borealis-4B-instruct-preview: benchmark results
NB-AiLab at the National Library of Norway publishes Borealis 4B Instruct Preview, a Norwegian instruction fine-tune of google/gemma-3-4b-it. This is the experimental Transformers checkpoint fine-tuned on textual instructions, distinct from its GGUF and MLX conversion releases. Provider: National Library of Norway. Released 2025-12-22. Access: Open.
Aggregate Rating 1478 ± 36, rank #2681 of 10916 rated models, from 47 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EuroEval Swedish NLU - Swerec | 78.93 | Matthews correlation coefficient (%) | 89.6 |
| EuroEval Faroese NLU - FoQA | 75.22 | Character F1 (%) | 89 |
| EuroEval Icelandic NLU - NQII | 54.82 | Character F1 (%) | 84.3 |
| EuroEval Norwegian NLU - Norec | 57.93 | Matthews correlation coefficient (%) | 82.2 |
| EuroEval Norwegian NLU - Norquad | 70.59 | Character F1 (%) | 78.9 |
| EuroEval Swedish NLU - Multi Wiki QA SV | 72.53 | Character F1 (%) | 69.6 |
| EuroEval Faroese NLU - FoSent | 45.37 | Matthews correlation coefficient (%) | 69.2 |
| EuroEval Icelandic NLU - MIM-GOLD NER | 56.53 | Micro F1 excluding MISC (%) | 68.3 |
| EuroEval Faroese | 44.17 | Average Score (%) | 66.3 |
| EuroEval Icelandic Common Sense Reasoning | 24.8 | Common Sense Reasoning Average Score (%) | 65 |
| EuroEval Icelandic NLU - Hotter and Colder Sentiment | 39.87 | Matthews correlation coefficient (%) | 63 |
| EuroEval Icelandic NLU | 37.8 | NLU Average Score (%) | 62.2 |
Interactive version: theaggregate.ai/model?slug=borealis-4b-instruct-preview · How It Works · Data refreshed daily, snapshot 2026-11-08.