Fourth HFMI Community Meeting in Berlin
On May 4–5, 2026, the fourth “HFMI Community Meeting” took place at MDC-BIMSB (Max Delbrück Center for Molecular Medicine in the Helmholtz Association) in Berlin. The event once again brought together the growing community of the Helmholtz Foundation Model Initiative (HFMI), following the previous meeting held in Cologne at “Wohngemeinschaft”.
What is HFMI?
The Helmholtz Foundation Model Initiative (HFMI) was launched in 2024 within the Helmholtz Association to specifically advance the development of foundation models for scientific research. The initiative began with four pilot projects and a central Synergy Unit that addresses cross-cutting challenges and strengthens collaboration across disciplines.
During 2024, a second call for proposals led to additional projects, bringing the total to seven pilot projects within the initiative. These span a broad spectrum - from health and life sciences to Earth and climate systems, as well as materials research and machine learning methods development. The Synergy Unit organizes community meetings twice a year to foster exchange, networking, and joint strategic development.
During the community meeting, it was reaffirmed that HFMI is more than a framework for AI-driven scientific projects; it is also a forum where people feel safe expressing their scientific concerns about AI and can build synergies with an energetic perspective looking into the future.
The Earth System Data Exploration (ESDE) group at the Jülich Supercomputing Centre (JSC) was also represented. Participants included Martin Schultz (CESOC Director & Group Lead) and Savvas Melidonis (ClimateAI Lead). Their participation highlights the close link between HFMI and the development of data-driven Earth and climate modeling, as well as high-performance computing in Europe.
All the projects have been developing remarkable capabilities: tools to unify data standards, identify missing metadata, create well-curated, large AI-ready datasets, establish robust pipelines and train increasingly convincing models. They are also building their own ecosystems. The HFMI is now entering the next phase: identifying and utilising synergies, improving models, and commencing test runs with user communities.
A key focus of the meeting was the topic of agentic AI - approaches, challenges, and opportunities (“Frontier AI” in Germany and Europe), explored in depth in a talk by Sebastian Lobentanzer (Helmholtz AI). Particularly engaging were discussions on the role of agentic AI approaches and the broader question of how foundation models will evolve and be positioned scientifically in the long term. It also became clear that development remains within a highly experimental and demanding landscape—especially regarding data, infrastructure, and the reproducibility of modern models.
HFMI’s positioning within the European context was further discussed with insights from Florian Grötsch, alongside contributions on the AI Pact Forum by Michael Schmuker. Potential collaborations and joint funding proposals were developed, followed by strategic discussions on the future of foundation models within Helmholtz and Europe.
Data is one of the central challenges for all members (curating good data, structuring it, exploiting metadata, loading it, transferring it); some disciplines are a bit ahead of others (climate simulation, earth observation) due to the nature of their data recording processes.
A central theme across all sessions was the role of data: its quality, structure, availability, and use as the foundation for robust models.
What stands out in particular is the strong community spirit that HFMI has enabled across all research disciplines.
A recurring impression from the meeting was the significantly strengthened cohesion within the initiative.
Conclusion
The fourth HFMI Community Meeting clearly demonstrated how the initiative is evolving from a set of individual pilot projects into a connected research community. Alongside methodological advances in foundation models, increasing attention is being paid to infrastructure, data strategy, and long-term scientific integration.
With the expansion of projects and a stronger European positioning, HFMI appears to be entering a new phase - characterised by collaboration, synergies, and the search for sustainable structures for AI in science.
More information: https://www.mdc-berlin.de/news/events/hfmi-spring-community-meeting