Arktis AI 1.0
An AI model developed in Norway for language, domains and workflows with high requirements for precision and control.
Purpose
Arktis AI is developed to solve concrete tasks in Norwegian and European organisations: understanding documents, retrieving relevant knowledge, structuring information and carrying out multi-step tasks through conversational and agent-based interfaces — under human control.
Why we are building it
General models are trained for breadth. Business-critical work requires precision in a specific language, a specific domain and a specific regulatory framework. We are not only building a good conversational partner, but infrastructure for digital work that can be used where requirements are strictest.
Problems the model addresses
The work targets tasks that today require extensive manual document handling.
- Preparing and summarising extensive case material.
- Retrieving knowledge in closed document collections.
- Structuring unstructured information.
- Quality assurance against internal requirements and regulation.
Languages
Development covers Norwegian, English, German, French and Spanish. Norwegian precision is the primary priority and forms the basis for the evaluation work.
Known limitations
The model is in development. The following applies until documented otherwise.
- No published benchmark results exist.
- Answers may contain errors and must be verified by a qualified person.
- The model must not be the sole basis for decisions, medical assessments or safety-critical actions.
- Capabilities marked «on the roadmap» are not available.
Training and adaptation method
The method description will be published as part of the model card with the first version. We do not publish details of data sources or training setup before they can be documented precisely.
Evaluation method
The evaluation framework describes category, dataset, method, date, model version, limitations and comparison baseline. No result is published without these fields.
Safety testing
Safety work covers input and output controls, access management and testing against misuse scenarios. Results are published in the trust centre when available.
Deployment
The model is intended to run as a managed API, in European or private cloud, and as an on-premise installation in the customer's own environment.
API status
The API is not publicly available. Developers can register for updates on the developer page.
Specification
- Model name
- Arktis AI 1.0
- Status
- In development
- Developed in
- Lierne, Norway
- Languages
- Norwegian, English, German, French, Spanish
- Deployment
- Managed API, European cloud, private cloud, on-premise
- Regulatory baseline
- EU AI Act, GDPR
- Model card
- Published with the first version
- Published evaluations
- None yet
Version history
- 1.0In developmentFirst model version. Model card, evaluation framework and limitations will be published at the same time.
Questions about the model
- Is the model available today?
- No. Arktis AI 1.0 is in development. Organisations can request early access and take part in the dialogue about use cases.
- Do you publish benchmarks?
- Not before the evaluation framework is published. Results without method, date and limitations are not documentation.
- Can the model run locally?
- On-premise installation is central to the product direction, particularly for the public sector and critical infrastructure.