NbAiLab/borealis2-31b-preview

VISIONPricing:Input $0.48 / Cached $0.1 / Output $1.44Concurrent Unit Cost:2Model Size:31BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 28, 2026License:otherArchitecture:Transformer Featherless Exclusive Cold

NbAiLab/borealis2-31b-preview is a 31-billion parameter instruction-tuned preview model developed by NbAiLab. It is based on unsloth/gemma-4-31B-it, continuously pre-trained on a 47-billion word dataset with a focus on Nordic languages, and fine-tuned on textual instructions. This model is designed for Norwegian-centric assistant tasks like drafting, summarization, Q&A, and light reasoning, and is intended for early evaluation of Norwegian language coverage and writing style assessment.

Loading preview...

Borealis 2 31B (Preview) Overview

NbAiLab/borealis2-31b-preview is a 31-billion parameter instruction-tuned model from NbAiLab, released as a preview for early testing and feedback. It is built upon unsloth/gemma-4-31B-it, further pre-trained on the extensive Aurora dataset, and then supervised fine-tuned using NbAiLab/aurora-sft-2606.

Key Characteristics

  • Architecture: Based on Gemma 4, with a 31B parameter count.
  • Training Data: Continuous pre-training utilized the Aurora dataset, comprising 47 billion whitespace-separated words with a knowledge cutoff of January 1st, 2025. This dataset includes a significant proportion of Norwegian (Bokmål and Nynorsk), Danish, Swedish, Icelandic, Faroese, Finnish, and various Sami languages, alongside English.
  • Fine-tuning: Supervised fine-tuning was performed for three epochs on textual instructions, with loss calculated only on assistant responses.
  • Language Focus: Primarily designed for Norwegian-centric tasks, with coverage for Norwegian (Bokmål, Nynorsk) and English.

Intended Use Cases

  • Norwegian Assistant Tasks: Suitable for drafting, summarization, question-answering, and light reasoning in Norwegian.
  • Writing Style Assessment: Can be used to evaluate Norwegian writing style and quality.
  • Early Evaluation: Ideal for assessing model behavior, language coverage, and overall quality in a preview context.

Important Considerations

As a preview model, Borealis 2 31B is not fully safety-aligned and may produce harmful or biased content. It is not recommended for safety-critical applications and requires additional safety mitigations if deployed. Outputs may be unstable and prone to hallucinations.

Finetunes
1 models