Germany Launches Kolibri, a New AI Model Designed to Keep Data Under European Control

Date: October 5, 2026

Germany’s Aleph Alpha has released a new artificial intelligence model designed specifically for organizations that want greater control over their AI systems.

The model, called Kolibri, is an open-weight German-English model with 78 billion total parameters. It was developed and trained in Europe and is aimed at mission-critical applications in government and industry.

The launch is another sign that European AI companies are increasingly focusing on a concept that has become central to the continent’s technology strategy: AI sovereignty.

Rather than simply competing with American companies on chatbot performance, European developers are increasingly emphasizing control, transparency, data protection and the ability to run AI on infrastructure controlled by customers.

A Model Built for Germany

Kolibri has been developed by Heidelberg-based Aleph Alpha.

The company says the model is specifically optimized for German and English. Around 23% of its pre-training data is German-language material, while its tokenizer has also been optimized for German.

That may sound like a relatively small technical detail.

It is not.

Language efficiency can matter considerably when AI is deployed at large scale. German words can require different tokenization than English words, and inefficient tokenization can increase computational requirements.

Aleph Alpha therefore designed Kolibri with German-language workloads in mind.

78 Billion Parameters — But Only a Fraction Are Active

Kolibri uses a Mixture-of-Experts architecture.

The model contains around 78 billion parameters in total, but only roughly 3 billion parameters are active for each token.

This approach allows a large model to provide substantial capacity without requiring the entire network to be active for every part of every response.

For organizations operating AI infrastructure themselves, efficiency matters.

Running a massive model can require enormous amounts of computing power.

A more efficient architecture can potentially reduce those requirements.

Designed for Sensitive Organizations

Aleph Alpha is not positioning Kolibri primarily as a consumer chatbot.

Instead, the company is targeting public administration and industry.

That includes organizations dealing with sensitive documents, internal data and regulated information.

The model supports agentic workflows, retrieval-augmented generation and native tool calling.

Those capabilities are particularly relevant to businesses.

An AI system could, for example, retrieve information from internal documents and then use connected tools to perform a task.

That turns the model from a simple question-and-answer system into a component of a broader enterprise workflow.

Europe Wants Control Over AI

The launch comes as European policymakers are increasingly concerned about dependence on foreign AI infrastructure.

The European Commission’s AI Board has been discussing Europe’s own AI capabilities, frontier technologies, AI safety and the implementation of the EU AI Act.

European Central Bank President Christine Lagarde has also warned that Europe needs to become a producer of AI technology rather than relying primarily on imported systems.

That makes companies such as Aleph Alpha strategically interesting.

A European government does not necessarily want every sensitive document processed through infrastructure controlled by a company outside Europe.

In some cases, organizations may want to run models on their own infrastructure.

Kolibri is designed with that scenario in mind.

The Model Is Open Weight

Aleph Alpha has released the full model weights through Hugging Face under the Apache 2.0 license.

That gives developers considerably more flexibility than using a closed API.

Organizations can potentially deploy the model themselves and inspect its technical characteristics.

For European companies, that can be important for compliance and procurement.

Instead of simply trusting a black-box service, organizations can have greater control over where the model runs and how it interacts with their data.

Aleph Alpha Is Going After a Different Market

The company is not trying to beat every American AI laboratory at every possible task.

Instead, it is pursuing a European niche.

That niche consists of organizations that care deeply about data control, regulatory requirements, transparency and European infrastructure.

This could become a significant market.

Banks, governments, manufacturers, healthcare organizations and large enterprises all have reasons to keep sensitive information under tighter control.

The more AI becomes embedded in those organizations, the more important infrastructure sovereignty becomes.