
Community collective · Sankt Augustin, Germany
OpenGPT-X
German publicly funded consortium that trained the open Teuken-7B models, built from scratch for all 24 official languages of the European Union.
Sankt Augustin, Germany
- Founded
- 2022
- Type
- Community collective
- Based in
- Sankt Augustin, Germany
- Key papers
- 2
- Featured projects
- 8
About
OpenGPT-X was a collaborative project funded by the German Federal Ministry for Economic Affairs and Climate Action from January 2022 to December 2024. It brought together partners from industry, research and media, with Fraunhofer IAIS and IIS leading and compute from the Jülich Supercomputing Centre.
Its main output was Teuken-7B, a seven-billion-parameter model trained from scratch on the 24 official EU languages, with about half of its pretraining data not in English. Instruct and base versions appeared on Hugging Face from September 2024, and the project issued its public press release on 26 November 2024.
The team also published a European LLM Leaderboard in July 2024 and an LLM Workbook, and Deutsche Telekom's T-Systems became the first provider to move Teuken from research to commercial use.
Mission and ambitionsOpenGPT-X set out to develop versatile, trustworthy, multilingual and open language models made in Germany and to support their adoption in the German economy.
Known for
- 01
Teuken-7B multilingual European model
- 02
24 EU languages trained from scratch
- 03
European LLM Leaderboard
- 04
Publicly funded open model project
Milestones
- 2022
Project start
The consortium began work in January 2022 with federal funding.
- 2024
European LLM Leaderboard
The team published a leaderboard for European-language model evaluation in July 2024.
- 2024
Teuken-7B research releases
Instruct versions v0.4 appeared on Hugging Face in September and October 2024.
- 2024
Teuken-7B public release
Fraunhofer announced the open release of Teuken-7B on 26 November 2024.
- 2025
Teuken-7B v0.6
Updated base and instruct versions published between December 2024 and July 2025.
Open source
Fork of the few-shot evaluation framework (13 stars)
Data quality tooling (11 stars)
Fork for training transformers at scale (4 stars)
Project repository (2 stars)
Instruct model for research use, the most downloaded OpenGPT-X model.
Instruct model with commercial-use licence.
Instruct model, version 0.6.
- openGPT-X/leaderboard_datadataset
Leaderboard data for multilingual evaluation.
Research papers
Notable releases
Openness
Teuken weights are public on Hugging Face; some variants have research-only licences.
People
- Nicolas Flores-HerrProject lead, Fraunhofer IAIS
Identified as OpenGPT-X project lead and co-author of the Teuken-7B paper.
- Georg RehmResearcher, DFKI
Co-author of the Teuken-7B paper and speaker at the 2024 OpenGPT-X Forum.
- Stefan KesselheimResearcher, Jülich Supercomputing Centre
Co-author of the Teuken-7B paper; Jülich supplied the training compute.
- Mehdi AliLead author, Teuken-7B paper
First author of the Teuken-7B technical paper.
Listed from the lab's own pages and public profiles.
Contact and community
The project's contact page names the Fraunhofer IAIS address for questions about the project.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



