Soofi S marks the start of the project’s first release phase. The model is intended for use particularly in scenarios where organizations want to operate AI applications in a transparent and customizable manner on their own or sovereign infrastructure—such as in industrial processes, the analysis of extensive technical and regulatory documents, code generation, or agent-based AI systems. Trained from scratch on 27 trillion (27T) tokens, Soofi S—a 30-billion-parameter (30B A3B) mixture-of-experts model—combines high throughput with low energy consumption thanks to its hybrid Mamba-Transformer architecture.
Soofi S was primarily trained on English and German texts and achieves top scores among open-source models in its size class for English; in German, it leads its peer group. Soofi S is initially being released as a base model that can already be fine-tuned for specific domains; followed by fine-tuned variants for dialogue and agent applications.
“Whoever controls the base models controls a central part of future digital value creation and increases their sovereignty and resilience. With Soofi, we are creating an open foundation on which companies, small and medium-sized enterprises, and the public sector can develop transparent AI applications based on their own data without becoming permanently dependent on individual non-European models,” says Jörg Bienert, Managing Director of the Center for Sovereign AI at the German AI Association.
There is a particular focus on transparency: The consortium will not only provide model weights but also publish technical documentation on training methodology, data preparation, and the data pipelines used. This will make Soofi S more verifiable for companies, government agencies, and researchers, as well as adaptable to specific application areas. Soofi S and subsequent models are trained on Telekom’s Industrial AI Cloud in Munich using Nvidia’s open-source AI framework. Initial results show that the Soofi S base model can keep pace with—or even outperform—international models of comparable size in German and English benchmarks.
“Soofi S is not intended as just another general-purpose chat model, but rather as a technical foundation for industrial AI. Crucially, Soofi S not only performs well in benchmarks but can also be deployed reliably, efficiently, and transparently in production,” says Nicolas Flores-Herr, technical project manager for Soofi and team lead at Fraunhofer IAIS.
About the Project Partners
The Soofi consortium brings together research institutions, universities, and AI companies from Germany. The project is coordinated by the German AI Association.
Participants include:
- Fraunhofer IAIS: Dr. Nicolas Flores-Herr, Dr. Mehdi Ali, Dr. Michael Fromm, Dr. Max Lübbering
- Fraunhofer IIS: Jan Plogsties, Dr. Viktor Hangya, Dr. Lucas Weber
- German Research Center for Artificial Intelligence: Prof. Dr. Antonio Krüger, Prof. Dr. Philipp Slusallek, Dr. Daniel Porta, Dr. Simon Ostermann
- Julius Maximilian University of Würzburg / CAIDAS: Prof. Dr. Andreas Hotho, Jan Pfister, Julia Wunderle
- Leibniz University Hannover / L3S Research Center: Prof. Dr. Wolfgang Nejdl, Dr. Simon Gottschalk
- Technical University of Darmstadt/hessian.AI: Prof. Dr. Kristian Kersting, Ruben Härle, Lukas Helff, Maurice Kraise, Sebastian Sztwiertnia
- Berlin University of Applied Sciences: Prof. Dr. Alexander Löser, Tom Röhr
- Ellamind: Dr. Jan Philipp Harries, Björn Plüster, Maximilian Idahl
- Merantix Momentum: Dr. Stefan Dietzel, Dr. Patrick Putzky, Dr. Martin Genzel
The model is being tested in real-world application scenarios in collaboration with companies from the industry. The goal is to gather early-stage experience from real-world deployment contexts and to align the model’s further development closely with specific industry requirements, so that companies can integrate the model into their processes and products. If you are interested in testing or collaborating, please contact us at contact@soofi.info.
Fact Box – Soofi S at a Glance
- Goal: A sovereign, transparent AI base model for industrial applications
- Target groups: Companies, SMEs, government agencies, research institutions, and startups
- Focus: Traceability, adaptability, operation on sovereign infrastructure
- Technical details: Published in the Tech Report
- Project: Soofi – Sovereign Open Source Foundation Models (www.soofi.info)
- Funding: Federal Ministry for Economic Affairs and Energy, financed by the
European Union (NextGeneration EU)
- Coordination: German AI Association, Jörg Bienert
About Soofi
Soofi – Sovereign Open Source Foundation Models – is a German consortium project with a European focus aimed at developing sovereign AI foundation models. The goal is to provide powerful, transparent, and openly accessible foundation models for industry.
– – – – –
Further Links
👉 www.iais.fraunhofer.de
Photo: Soofi
