Teuken-7B
Multilingual LLM for all 24 official EU languages
Teuken-7B is a 7 billion parameter language model built by German research institutes under the OpenGPT-X project. Released in 2024, it supports all 24 official European Union languages and aims to reflect European values better than English-centric alternatives.
- Supports all 24 official EU languages
- Trained on 4 trillion tokens
- Released under Apache 2.0 license
- Developed by German research institutes
- Not designed for math or coding tasks
What it is
Teuken-7B is a large language model with 7 billion parameters designed to handle all 24 official languages of the European Union. It was developed by a consortium of German research bodies including Fraunhofer, Forschungszentrum Jülich, TU Dresden, and DFKI. The model forms part of the OpenGPT-X initiative, which received funding from the German Federal Ministry of Economics and Climate Protection. Its primary goal is to provide stable performance across European languages and align outputs with European values, addressing the gap left by models that focus heavily on English.
How it came about
The model was introduced in September 2024 through a research paper titled Teuken-7B-Base & Teuken-7B-Instruct: Towards European LLMs. The development team included over forty authors from various German institutions. The project aimed to create a model that embraces linguistic diversity within Europe. The base version and an instruction-tuned variant were released to support both commercial and research applications. The data used for training had a cutoff date of September 2023.
Inside the model
This decoder-only transformer model was pre-trained on 4 trillion tokens from public sources. Approximately 60 percent of this data was non-English. The developers used a custom multilingual tokenizer to optimize performance across different languages. For instruction tuning, they combined English, German, and multilingual datasets. German data was partly created by translating English instructions using another model, while excluding code snippets to avoid errors. The English portion was filtered for quality using reward scores from a separate ranking model.
What it can do
Teuken-7B is specialized for multilingual tasks across the European Union. It shows strong results on benchmarks such as European versions of ARC, HellaSwag, and TruthfulQA. The creators state it offers more stable results in EU languages than English-centric models. However, it is not intended for mathematics or coding tasks. Like other large models, it may still produce biased or toxic content despite filtering efforts, and it is not free from hallucinations.
How to use it
The instruction-tuned version is available under the Apache 2.0 license, allowing commercial and research use. Users can download the weights from Hugging Face. The base model is available only on request via email to the project team. Running the model requires specific Python libraries and adherence to a provided prompt template. It can be deployed locally or served using tools like vLLM, with options to set the chat template language.
Safety and testing
The training dataset was filtered to reduce inappropriate or harmful outputs. Despite these measures, the developers warn that the model may still generate biased, toxic, or hallucinated text due to the scale and nature of the data. They explicitly state that the model is not completely free from these risks. Users are advised to be aware of these limitations when deploying the system.
Questions people ask about Teuken-7B
Which languages does Teuken-7B support?
It supports all 24 official languages of the European Union, including Bulgarian, Czech, Danish, German, Greek, English, Spanish, Estonian, Finnish, French, Irish, Croatian, Hungarian, Italian, Lithuanian, Latvian, Maltese, Dutch, Polish, Portuguese, Romanian, Slovak, Slovenian, and Swedish.
Is Teuken-7B free to use commercially?
Yes, the instruction-tuned version is released under the Apache 2.0 license, which permits both commercial and research use. The base model is available only upon request from the project team.
Can I use Teuken-7B for coding tasks?
No, the developers state that the model is not intended for use in mathematics or coding tasks. It is specialized for general multilingual natural language processing across European languages.
Good reading and sources
There is no English Wikipedia article on Teuken-7B yet, so the facts on this page come from the sources listed above: huggingface.co/openGPT-X/Teuken-7B-instruct-commercial-v0.4, arxiv.org/abs/2410.03730. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryLLM wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.