The Soofi project, launched in November 2025 with the participation of L3S, has reached its first major milestone with the development of its language model, Soofi S.
Soofi S can understand language and generating text autonomously. It can be deployed in chatbots or support industrial applications by analysing large document collections, generating programming code, or assisting with process analysis. Technically, the model is based on a Mixture-of-Experts architecture and has been trained using 30 billion parameters on around 27 trillion text tokens.
As part of the project, researchers at L3S are working to ensure the safety of the language models being developed. Their goal is to design systems that do not generate harmful content, such as offensive or discriminatory statements. Dangerous instructions or the disclosure of confidential information is also strictly prohibited. In addition, L3S is investigating the models’ cultural understanding, examining the extent to which they reflect European and German values. The research centre also coordinates collaboration with Deutsche Telekom’s Industrial AI Cloud in Munich, where Soofi S was developed, and contributes its expertise throughout the entire development process.
“Soofi demonstrates how essential excellent research and highly skilled professionals are for Europe’s technological sovereignty. The demand for qualified AI experts will continue to grow in this forward-looking field. At the same time, the project highlights that innovation in AI extends far beyond algorithm development and depends on powerful hardware, modern communication systems, and intelligent technical infrastructures,” says Prof. Dr.-Ing. Peter Werle, Dean of the Faculty of Electrical Engineering and Computer Science at Leibniz University Hannover.
A key objective of the Soofi project is to strengthen Europe’s technological sovereignty. With Soofi S and future model versions, the consortium is creating an independent alternative to existing commercial language models developed by international providers. The project also places a strong emphasis on transparency, including comprehensive documentation of the data used and the methods applied to prepare it.
“At a time when access to powerful AI systems in Europe can no longer be taken for granted, developing our own European solutions is becoming increasingly important. With Soofi, we are making a significant contribution to sovereign and transparent AI development in Europe,” says Prof. Dr. techn. Wolfgang Nejdl, Director of the L3S Research Center.
Soofi S is now entering a testing phase together with industrial partners in realistic application scenarios. The aim is to gather early insights from real-world use cases and further develop the model in line with industry requirements. This will enable companies to integrate Soofi language models into their products and business processes in the future.
Consortium and Funding
The Soofi consortium brings together six leading German research institutions, including L3S, Fraunhofer IAIS, Fraunhofer IIS, the German Research Center for Artificial Intelligence (DFKI), the University of Würzburg, TU Darmstadt, and Berlin University of Applied Sciences and Technology, as well as two start-ups. The project is coordinated by the German AI Association (KI Bundesverband) and is funded by the German Federal Ministry for Economic Affairs and Energy with approximately €20 million.
