eurollm.io

What It Is

EuroLLM is a European-led open-source large language model project that aims to build a native multilingual LLM supporting all 24 official languages of the European Union. The project involves institutions such as Instituto Superior Técnico, the University of Edinburgh, Université Paris-Saclay, Unbabel, and Naver Labs, with support from Horizon Europe, ERC, EuroHPC, and others. Its positioning is closer to a foundation model and research infrastructure than a ready-to-use SaaS application.

Core Capabilities

The current flagship model is EuroLLM-22B, with 22B parameters. It was trained on more than 4 trillion tokens across 35 languages, supports a 32K context window, and is available in both Instruct and Base versions, targeting chat/instruction following and downstream fine-tuning respectively. EuroLLM-9B and EuroLLM-1.7B are also available, with the 1.7B model better suited for edge devices. The official website also previews the EuroVLM-9B vision-language model and the EuroMoE sparse expert model, but multimodal, vision, and speech capabilities have not yet been formally released.

Pricing and Integration

The project emphasizes Open Source in its public materials, stating that it is freely available for researchers, organizations, and European citizens, with model access provided via Hugging Face. However, it does not disclose details on commercial APIs, hosted inference, enterprise editions, SLAs, payment methods, or specific licensing terms. As a result, its low-cost advantage is clear, but users need to handle deployment, inference resources, monitoring, and engineering integration on their own.

Pros and Cons

Its strengths include clear multilingual coverage, making it especially suitable for EU-language Q&A, summarization, translation, and NLP research. It also offers multiple parameter sizes and both Base/Instruct versions, which helps with fine-tuning and application development. The limitations are that the official website does not provide a complete benchmark table, privacy policy, API documentation, or service support information. Its Chinese capability is not clearly stated, and Chinese-language scenarios are not its primary focus.

Who It’s For and Access from China

EuroLLM is suitable for universities, research institutions, application teams targeting the European market, and developers who want to build multilingual capabilities on top of open-source models. Access from China is not discussed in the source materials. Since the models are available on Hugging Face, actual access may depend on the local network environment. If the focus is Chinese-language applications, alternatives such as Qwen, Llama, Mistral, and Gemma can also be evaluated.