Built for the Norwegian language, European data sovereignty and enterprise control.
Od1n is built on a modern Transformer architecture and trained with a focus on stability, scalability and linguistic quality. This is not a fine-tuned version of an existing model – Od1n is trained from scratch on Norwegian and Nordic data.
Od1n V5 – sovereign Norwegian base model
Od1n (3B) measures itself against the largest open language models on Norwegian – and beats models up to 3× its size. The two NorMistral models score higher overall, but they are 2–4× larger; Od1n beats every international model, and goes past NorMistral-7B on Norwegian public administration.
Od1n (3B) beats NorMistral-7B – a model more than twice its size – on Norwegian public administration.
better Norwegian score (lower bits-per-byte) than Qwen2.5-3B, which is the same size.
Beats Gemma-9B, Llama-8B, Mistral-7B and Qwen-3B – all equal or larger.
The Norwegian score is byte-invariant BPB (bits-per-byte) – lower is better. Being byte-invariant, it is tokenizer-independent and makes cross-model comparison fair. All models were run with the same evaluation code and context (2048 tokens) on a fresh Norwegian 2026 corpus (parliament, municipal administration, Nynorsk and code) – published after the competitor models' training cut-off, so it is contamination-free. Competitors are evaluated as finished, published models; Od1n is measured at completed pre-training. These are internal measurements and are not third-party verified.
Talk to us about how a sovereign Norwegian language model can be adapted to your organisation.
Contact usSee Od1n V5