There is a lot of talk at the moment about Large Language Models (LLMs) and their potential impact on all kinds of things, from education to the economy. However, much of the development and focus of these models has been on English, often leaving other languages, especially the smaller languages of Europe, behind. This can lead to inaccuracies and a lack of cultural nuance. But now, a new development from Europe is aiming to change that.

Tilde, a winner of the European AI Grand Challenge, has released TildeOpen LLM, a powerful 30-billion parameter language model specifically optimized for European languages. This is a significant step forward for the European AI ecosystem, demonstrating a growing capacity to develop world-class AI infrastructure. The model was trained using 2 million GPU hours on the EuroHPC LUMI supercomputer, a remarkable achievement in under a year.

One of the most significant aspects of TildeOpen is its focus on being "built for Europe." With 24 official languages and over 60 regional languages in the EU, the dominance of English-centric models is a real issue. As Tilde CEO Artūrs Vasiļevskis explains: “Popular commercial language models, such as ChatGPT, are mostly trained using English language data, implying that the results generated in English will be of better quality than those generated in other, less common languages. This happens to lead to awkward sentence structures and word order, grammatical errors or even inaccurately used and translated terms.”

TildeOpen aims to address this by providing equal support for all official EU languages, with a particular focus on those that are often underrepresented in current Large Langauge Model solutions, such as the languages of the Baltic countries, Ukrainian, and Turkish. This is not just a matter of technical capability, but also of cultural and linguistic diversity. More than 200 million Europeans, nearly half of Europe’s population, speak these so-called small languages.

Beyond its multilingual capabilities, TildeOpen is built on a foundation of European values. It is an open-source model, promoting transparency and ethical data handling. This openness allows for greater scrutiny and customization. Developers can adapt the model for national, legal, or sector-specific needs, from building AI assistants to creating specialized services for public administration, education, or local companies.

Security and data sovereignty are also at the core of the TildeOpen project. Unlike many popular commercial models hosted in the US or Asia, TildeOpen can be deployed locally or in trusted European cloud environments. This gives organizations full control over their data, ensuring compliance with EU data protection and privacy standards. Vasiļevskis emphasizes that this is a key requirement of the European Commission, which wants to see EU developers creating AI products for the internal market that are stored in secure, compliant resources within Europe.

The model is also designed to be efficient and sustainable, with more efficient tokenization that leads to lower costs and a reduced carbon footprint. This is an important consideration as the use of AI continues to grow.

TildeOpen LLM is more than just a new piece of technology; it represents a commitment to a more inclusive and diverse AI landscape in Europe. By providing a powerful, open, and secure multilingual base, it empowers developers and organizations across the continent to build AI solutions that are tailored to their specific linguistic and cultural contexts.

The first version of TildeOpen LLM has been released on the Hugging Face platform, making it freely accessible to researchers, companies, and anyone interested in building with it.

References

•Tilde. (2025, September 3). Tilde releases artificial intelligence model for the European languages – TildeOpen LLM.

About the Image

The image explores the hidden mechanics behind recommendation systems on social media. Have you ever wondered how platforms seem to know exactly what you're interested in, like that carpet you're considering buying? And what about the vast amounts of personal data we unknowingly leave behind while navigating our online lives? The image visualises the invisible algorithms that shape our online experiences, highlighting how our behaviours are tracked and used to predict our desires and choices.