European Commission Launches Institutional LLM and Benchmark for EU Languages

The European Commission’s Directorate-General for Translation (DG Translation) has released an open large language model (LLM) trained with multilingual data from the EU institutions.

Announced on July 16, 2026, the EU Institutional LLM is designed to improve support across all 24 official EU languages and better handle EU terminology and institutional content. DG Translation described the model as “a better option for EU topics and more suitable for EU public administrations, small businesses, academia and non-governmental organisations.”

The release includes a base model and a version adapted to follow instructions. Both build on an open model (Mixtral 8x7B) from Mistral AI, selected as part of the Commission’s focus on European technology.

The model was developed using Euramis, the EU institutions’ database of professionally translated legislative, administrative, and policy texts. According to DG Translation, the EU Institutional LLM outperformed the original Mixtral model across every language tested on a benchmark based on EU institutional texts.

The models are released under the Apache 2.0 license, a permissive license that allows them to be reused and modified. However, access is limited to legal entities based in the EU.

The EU Institutional LLM already powers eSummary, the Commission’s multilingual document summarization service, while future iterations will be used in more of its AI-based language tools.

The European Commission positioned the EU Institutional LLM as part of its push to strengthen European sovereign AI. Earlier in July, the Commission selected the EUROPA consortium to develop a separate open-source model covering all 24 official EU languages.

2026 Slator Market Report: Language Solutions & AI

The 130-page Slator Report maps a USD 30.85 billion global market shaped by multilingual AI, enterprise AI operationalization, and the convergence of language technology, media, accessibility, and real-time communication.

$995 BUY NOWIncluded in our Growth, Pro, and
Enterprise plans. Subscribe now!

DG Translation Develops EU Multilingual AI Benchmark

Following the model release, DG Translation announced EU MMLU, a new benchmark for evaluating LLMs across EU languages.

“Building a model that covers all EU official languages is only part of the challenge,” DG Translation said, adding that reliable tools are also needed to measure model performance across languages.

EU MMLU is based on Massive Multitask Language Understanding (MMLU), a widely used benchmark that tests models using multiple-choice questions across different subject areas. 

DG Translation said many existing multilingual benchmarks are created mainly through AI translation. For EU MMLU, professional translators from DG Translation and students from the European Master’s in Translation network translated the questions without using AI translation as a starting point and revised one another’s work. Student project managers coordinated the workflow, while DG Translation’s language units provided guidance on terminology and language-specific issues.

The first release includes more than 1,000 questions and lists 16 official EU languages. DG Translation plans to extend it to the remaining official languages.

EU MMLU is available under a Creative Commons Attribution 4.0 license — which permits reuse and adaptation as long as the source is credited — through Hugging Face and the European Language Data Space.

DG Translation also published a set of quality criteria for developing EU-oriented multilingual benchmarks, with the longer-term goal of establishing “a standard for multilingual AI evaluation in Europe.”