Latvian AI company Tilde has launched TildeOpen LLM, an open-source large language model (LLM) trained on Europe's flagship supercomputer, LUMI. With 30 billion parameters, TildeOpen is designed to excel in European languages—covering 34 languages including 24 official EU ones plus Ukrainian, Norwegian, Icelandic, Turkish, and Balkan languages—addressing the underserved markets often neglected by global AI models.
TildeOpen LLM is an open-source, 30B-parameter model trained on EU supercomputers LUMI and Jupiter.
Specialized for European languages, offering top performance in both major and smaller regional languages.
Outperforms open models like Gemma-27B and EuroLLM-22B on benchmarks like Belebele reading comprehension.
Integrates safeguards against disinformation, filtering Kremlin-aligned propaganda in training data.
Supports local deployment or trusted European clouds, complying with GDPR, AI Act, and data sovereignty laws.
Developed with support from the European AI Grand Challenge with 2 million GPU hours granted.
TildeOpen LLM addresses a critical gap in AI by offering high-quality multilingual capabilities for languages often overlooked by English-centric models. Its training on the largest European supercomputers ensures efficient, secure, and accurate AI for wider European audiences.
The model incorporates filtering mechanisms developed in collaboration with media monitoring authorities to block disinformation from hostile actors, notably Russian state-aligned networks, ensuring trustworthy model behavior and data integrity.
Released under a CC-BY-4.0 license on Hugging Face, TildeOpen facilitates customization by researchers, universities, startups, and enterprises, encouraging innovation while adhering to strict European privacy and security standards.
Tilde is a European AI-driven language technology company that advances language equality and trustworthy AI solutions across the continent. With expertise in multilingual datasets and cutting-edge research, Tilde develops AI products compliant with EU governance, fostering innovation while preserving linguistic diversity.