Telekom debuts language model 'Made in Germany' from OpenGPT-X

With the EU project OpenGPT-X, funded by the Federal Ministry for Economic Affairs and Climate Protection (BMWK), Germany is driving forward the development of open and trustworthy AI language models. This strengthens digital sovereignty and gives German authorities and institutions in particular access to pioneering European generative artificial intelligence (GenAI) technology.

A consortium led by the Fraunhofer Institutes IAIS and IIS developed the Teuken-7B language model with other partners such as the TU Dresden, the German Research Center for Artificial Intelligence GmbH (DFKI) and the Jülich Research Center. The language model, trained on the Jülich supercomputer JUWELS, has been available under an open source license since November 26. 

Deutsche Telekom successfully transferred the development of the European language model from research to industry. This makes the telecommunications company the first provider to provide access to Teuken-7B commercially. Authorities and companies can now use a powerful, trustworthy and open source alternative to the major common models. Telekom is thus making a significant contribution to Germany's digital sovereignty. 

"When, if not now, does Europe need more sovereign solutions? Only through joint European efforts can we create competitive alternatives to the large international providers. Teuken-7B from OpenGPT-X is a first successful project 'Made in Germany'. It is an important piece of the puzzle in the digital strategy of customers for whom sovereignty is the be-all and end-all," said Dr. Ferri Abolhassan, CEO of T-Systems and member of the Board of Management of Deutsche Telekom AG.  

Responsible AI with the quality seal “Made in Germany” 

When developing large AI language models, OpenGPT-X relies on "Made in Germany". The needs of both European companies and research are taken into account. This enables new data-based business solutions for the future. The models are characterized by multilingualism, openness (open source) and trustworthiness and comply with Deutsche Telekom's ethical and responsible AI principles. The Teuken-7B model was trained multilingually from scratch and, thanks to a specially developed tokenizer, offers high energy and cost efficiency for applications in all 24 official European languages. With its comprehensive open source licensing (Apache 2.0) and optimized performance, it sets new standards for AI solutions in Europe.  

"Providing Teuken as an open source model has several advantages: Companies can adapt the model to their needs - i.e. develop specialized applications with it. They can also decide whether they want to run the model locally on their own infrastructure or with a trusted cloud provider of their choice. If desired, sensitive data can therefore remain in the company." - Dr. Nicolas Flores-Herr, project manager at Fraunhofer IAIS.

Diverse application possibilities with flexible operating models

Telekom offers a variety of advantages and possible uses of Teuken-7B, tailored to the specific requirements of European companies and authorities. This strengthens the freedom of choice of German companies and public clients in the growing market for generative AI. With AI Foundation Services, the Telekom subsidiary T-Systems provides flexible operating models for this. Companies can choose whether they want to use the models in highly secure, certified Telekom data centers in Germany, GDPR-compliant in Europe, or even dedicated to their own infrastructure. Telekom offers solutions that meet the highest standards of security and compliance, especially for regulated industries and data that requires particular protection.  

The new language model "Teuken-7B" is also directly integrated into the standardized Telekom product Business GPT. It supports "out-of-the-box" RAG applications (retrieval augmented generation) and summarizes documents and internal company chatbots. Using a uniform API (network interface), companies can also integrate the model into existing AI assistants, agents, and their own chatbots using AI Foundation Services. The AI ​​Foundation Services also enable the development and operation of generative AI applications on scalable and secure platforms.