Jülich Hosts EVE Language Model for Earth Observation and Earth Science - HPCwire

Off the Wire Press Releases

Nov. 3, 2026 — The Jülich Supercomputing Centre (JSC) has started a new collaboration with Pi School, an international centre of excellence for AI headquartered in Rome that brings together innovators, thinkers and entrepreneurs. As a result, JSC’s Helmholtz Blablador infrastructure now hosts the EVE-Instruct model, a large language model (LLM) for Earth observation (EO) and Earth system science, where EVE stands for Earth Virtual Expert.

EVE is funded by the European Space Agency (ESA) Φ-lab and developed by Pi School and other collaborators. EVE is built on an open-source foundation model and fine-tuned on curated EO and Earth science corpora. It combines structured citations and hallucination-aware evaluation, so its answers are grounded in and traceable to the underlying sources. EVE lets users query scientific literature, institutional content and technical documentation in natural language, which supports reliable Earth Intelligence applications.

Running the EVE-Instruct model on Helmholtz Blablador means the model is served from European public research infrastructure through an inference service offered by the JUPITER AI Factory (JAIF), which is part of the EuroHPC AI Factories network. Together with its open data, model weights and code, this makes the whole EVE stack open, from training to deployment. JSC’s high-performance computing resources and JAIF’s inference services also provide a foundation for scaling EVE as its user base and agentic workloads grow.

Since October 2026, EVE has been publicly available to everyone. To interact with the model, visit the EVE website and click on ‘Try EVE!’ to create a free account. Developers can connect to EVE via an OpenAI-compatible API. The models, data and code are also openly available on Hugging Face and GitHub for organizations that want to run EVE on their own infrastructure.

Dr. Gabriele Cavallaro, head of JSC’s Simulation and Data Lab (SDL) AI and ML for Remote Sensing, said: “In June 2026, our FAST-EO project, funded by ESA Φ-lab, entered its second phase to further advance the state of the art in EO foundation models. JSC is a consortium partner alongside DLR, IBM Research, KP Labs and ETH Zurich. The timing is ideal, as the EVE project also entered its second phase in August 2026. This opens up new opportunities to strengthen collaboration between the two projects as EVE evolves into a multimodal, agentic AI platform that aims to integrate scientific foundation models, including FAST-EO’s TerraMind model, with cutting-edge supercomputing resources.”

Dr. Alexandre Strube, team leader at Helmholtz Blablador, said: “Blablador has been hosting scientific LLMs for years, and the jump in quality from the first ones to EVE is palpable. The second phase of EVE promises to be even more interesting. We welcome this model into our hosting platform!”

For more information about research and work activities of the SDL AI and ML for Remote Sensing at JSC, visit the SDL’s webpage.

Source: JSC