TLDR: French startup Dragon LLM, formerly Lingua Custodia, has announced the development of ‘Dragon,’ a groundbreaking generative AI architecture. This European-designed model is touted as a more energy-efficient and accessible alternative to existing US and Asian AI systems, capable of delivering comparable performance with significantly less computational power. The initiative aims to foster European technological sovereignty in artificial intelligence.
Dragon LLM, a French startup that originated as Lingua Custodia, has made a significant stride in the artificial intelligence landscape by unveiling ‘Dragon,’ what it claims to be Europe’s first generative AI architecture. This innovative model, announced on November 5, 2025, represents a pivotal moment for European technological independence in the rapidly evolving AI sector.
The ‘Dragon’ architecture is designed to be frugal, efficient, and energy-conscious, setting it apart from the widely adopted Transformer-based systems that power leading models from companies like OpenAI, Anthropic, and Mistral AI. According to Dragon LLM, their new architecture can achieve performance levels comparable to these established models while utilizing only one-third of the computational resources. This efficiency allows the ‘Dragon’ model to operate on conventional servers, negating the need for extensive GPU infrastructure, thereby lowering both operational costs and energy consumption.
Olivier Debeugny, CEO and founder of Dragon LLM, highlighted the strategic importance of this development. ‘The LLMs that power most generative AI use cases today are built using a certain architecture — the Transformer — which underpins about 95% of all models on the market,’ Debeugny stated. He further emphasized, ‘We were able to create and validate our demonstration model without massive private investment. It proves that Europe has the expertise and the computing infrastructure to innovate in foundational AI technologies, and to stand alongside the U.S. and China.’ This project was made possible after Dragon LLM was selected as a winner of the European Large AI Grand Challenge in June 2024, granting them access to high-performance computing (HPC) resources, including early access to the JUPITER supercomputer in Germany and Leonardo, both part of the EuroHPC initiative.
Also Read:
- EU Appoints Experts to Lead Generative AI Content Labeling Code of Practice
- OpenAI Launches Executive Education Program in Amsterdam to Foster AI Leadership
The company’s vision extends beyond mere technological advancement; it aims to foster a sovereign and sustainable AI ecosystem within Europe. By drastically reducing energy consumption and computing needs, Dragon LLM’s architecture could enable AI models to run locally on SME servers or even smartphones, thereby democratizing access to generative AI in environments that have traditionally relied on large-scale cloud infrastructure. The initial version, Dragon-3B-Base-alpha, built with 3.6 billion parameters, has already been made publicly available on the Hugging Face platform, with commercial versions anticipated in the coming months. This move positions Dragon LLM as a key player in building responsible and accessible AI solutions tailored for European businesses, reinforcing the continent’s capacity to innovate and create its own foundational AI technologies.
