German AI company Aleph Alpha has launched Kolibri, a specialised large language model (LLM) built for mission-critical applications in public administration and industry.
Referred to as sovereign AI made in Germany and released on 3 October 2026, Aleph Alpha says Kolibri is designed for organisations that require data control, documented compliance, and a model trained to operate natively in German.
“Kolibri demonstrates that we have the talent and expertise in Germany to develop competitive AI models. For us, AI sovereignty means freedom of choice by retaining the ability to build and advance this technology, and giving customers control over how they use it. Our ambition is to bring that capability into everyday operations in industry and public administration,” said Ilhan Scheer, CEO of Aleph Alpha.
Founded in 2019, Aleph Alpha is on a mission to build sovereign, human-centric AI. Along with Mistral, the company has become one of the two European champions challenging frontier labs in the US and China. It employs around 200 people across four locations in Germany, and says its specialised LLMs are used by enterprises and government institutions looking to maintain data sovereignty and build trustworthy AI applications.
Kolibri is a mixture-of-experts (MoE) model combining 78 billion total parameters with approximately 3 billion active per token. The model is designed in such a way that instead of activating the entire model during inference, its architecture selectively engages only the relevant parts, a design approach that improves efficiency without compromising on capability. Aleph Alpha says German-language text makes up around 23% of its pre-training data, and it has developed a German-optimised tokeniser that represents German text using fewer tokens to enable faster and more efficient processing of German content.
The model supports agentic workflows, retrieval-augmented generation (RAG), and native tool calling. In RAG applications, Kolibri is trained to answer questions based on supplied documents and abstain from responding when those documents don’t contain sufficient evidence. In other words, Aleph Alpha claims to have reduced hallucinations in high-stakes settings. It also allows users to adjust the balance between reasoning depth and response speed through built-in reasoning controls.
To meet the compliance and provenance demands of regulated industries and government institutions, Aleph Alpha has screened Kolibri against a blocklist of more than 4.5 million URLs, drawing on sources including the European Commission’s Piracy Watch List. It has also accessed every third-party dataset individually for licence terms, lawful sourcing and adherence to opt-outs. In a technical report, Aleph Alpha has detailed how Kolibri addresses copyright, data protection and EU AI Act requirements, while its licence terms and supporting documents are available on Hugging Face.
Samuel Weinbach, co-founder and co-Chief Research Officer, Aleph Alpha said, “We designed Kolibri around the needs of German-language applications, from a tokeniser optimised for German to training the model to reason in German. For our customers, this is an important step of making AI useful in everyday workflows.”
The new sovereign AI model, Kolibri, follows the company’s April announcement of an agreement with Canadian AI firm Cohere to build a joint transatlantic sovereign AI company. While the merger is subject to regulatory approval, the two companies officially signed a definitive business agreement last month, announcing that the combined entity will operate globally as Cohere with headquarters, R&D centres, and leadership roles in Canada and Germany.
With Kolibri now available and a merger in the pipeline, Aleph Alpha is not only showing that Europe can build its own large language models but do so without compromising on sovereignty.