Germany’s Aleph Alpha Unveils Kolibri: A Sovereign 78B AI Model
In a significant move towards European AI autonomy, German AI firm Aleph Alpha has launched Kolibri—a robust, open-weight large language model tailored for government institutions and regulated industries requiring complete data sovereignty. The launch, which took place on October 3, 2026, was strategically timed to align with German Reunification Day.
Kolibri is a Mixture-of-Experts (MoE) model with a staggering 78.1 billion total parameters, and approximately 3.46 billion active per token. It supports an extensive context window of up to 1,048,576 tokens and incorporates multi-level reasoning and tool-calling capabilities.
The model was trained on nearly 24 trillion tokens across GPU clusters located in Germany and Finland. Its full weights are accessible on Hugging Face under the liberal Apache 2.0 license.
What distinguishes Kolibri is its “sovereignty by design” philosophy. Unlike most frontier models that regard German as a secondary language, Kolibri was intentionally designed to reason natively in German. This makes it ideally suited for German public administration. It was developed in compliance with the EU AI Act, the GDPR, and the General-Purpose AI Code of Practice.
Aleph Alpha, a leading AI challenger in Europe alongside Mistral, is currently in the midst of a $20 billion merger with Canadian AI firm Cohere. With €600 million backing from retail giant Schwarz Group, the merger aims to establish a formidable transatlantic competitor to U.S. and Chinese AI dominance.
Source: EU Startups – October 5, 2026