At an event organized by the Latvian Association of Artificial Intelligence (MILA), Tilde presented the new European large language model TildeOpen LLM — the first artificial intelligence solution of this scale developed in Europe, and the only model trained equally across 34 European languages, including Latvian.
The model was created on behalf of the European Commission and trained using the LUMI supercomputer, which enabled extremely high computational capacity and achieved a level of linguistic quality previously unattainable with global, English-centric models.
During the event, it was particularly emphasized that TildeOpen LLM significantly improves the accuracy of the Latvian language, eliminating syntax, morphology, and terminology errors that have so far limited the practical use of AI in public administration and other fields requiring high language quality.
The European-developed model also ensures a much higher level of data security, as it can be deployed within the infrastructure of Latvia or other EU member states without transferring sensitive information outside Europe. This is a key prerequisite for the adoption of artificial intelligence in healthcare, social services, the justice sector, education, and other public services.
Tilde pointed out that this solution offers several important benefits for Latvia’s public sector. Since the model can also be deployed locally, it allows secure use of AI without transferring personal data or restricted information outside the country. The model ensures high-quality results in Latvian, which is especially important for the public sector, where precise language processing is a critical requirement.
TildeOpen serves as a foundation for developing a wide range of AI services and can be customized for various applications based on institutional needs. When tailored for specialized tasks, it can analyze extensive documents, generate summaries, and extract key requirements, simplifying the review of procurement, project, and policy documentation.
It can also operate as an internal AI assistant that searches for answers only within trusted institutional resources, helping employees quickly find information and navigate processes and regulatory documents. At the same time, the model can become an essential tool for developing modern e-services, helping citizens more easily find relevant services and automating the preparation and explanation of documents.
This approach ensures that solutions are reliable, secure, and adapted to the needs of Latvia’s public administration.
TildeOpen LLM is already freely available to anyone on the Hugging Face platform and is being adapted for practical applications such as machine translation, document analysis, and internal chatbots. A new version with a significantly extended context window will soon be released, enabling AI to process longer documents and more complex data.
MILA notes that this technology represents an important step toward strengthening Latvia’s digital sovereignty and linguistic sustainability, as, for the first time on a European scale, a solution has been created that understands the structure of our language, adapts to the national context, and provides a secure, reliable, and professional foundation for the use of artificial intelligence in public administration and business.
