Germany's Aleph Alpha Launches Kolibri: A Sovereign AI Model for Government

Jan de Vries · 2026-10-05

Germany's Aleph Alpha just dropped something big: Kolibri, a sovereign AI model built for the most demanding government and industry applications. If you're tracking EU Inc news or thinking about incorporating a startup in Europe, this is a signal you can't afford to ignore. ### What Exactly Is Kolibri? Kolibri is a specialized large language model (LLM) designed for mission-critical tasks in public administration and industry. Aleph Alpha calls it "sovereign AI made in Germany," and it was released on October 3, 2026. The model is built for organizations that need tight data control, documented compliance, and native German language operations. Ilhan Scheer, CEO of Aleph Alpha, put it simply: *"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."* ### Under the Hood: Efficiency Meets Power Kolibri uses a mixture-of-experts (MoE) architecture. That means it has 78 billion total parameters but only activates about 3 billion per token. Instead of firing up the whole model every time, it selectively engages only the relevant parts. The result? Better efficiency without sacrificing capability. Aleph Alpha also trained Kolibri with German in mind. About 23% of its pre-training data is German text, and it uses a German-optimized tokenizer that represents German words with fewer tokens. That speeds up processing and makes the model feel more natural for German-language tasks. ### Built for Real-World Workflows Kolibri isn't just a lab experiment. It supports agentic workflows, retrieval-augmented generation (RAG), and native tool calling. In RAG applications, it answers questions based on supplied documents and—crucially—abstains from responding when the documents don't contain enough evidence. That's a big deal for reducing hallucinations in high-stakes settings like government or regulated industries. Users can also adjust the balance between reasoning depth and response speed through built-in controls. So you can dial up accuracy when you need it, or dial it down for faster answers. ### Compliance and Provenance: No Shortcuts Aleph Alpha screened Kolibri against a blocklist of over 4.5 million URLs, including sources from the European Commission's Piracy Watch List. They also checked every third-party dataset for license terms, lawful sourcing, and opt-outs. A technical report details how Kolibri addresses copyright, data protection, and EU AI Act requirements. License terms and supporting docs are available on Hugging Face. ### Why This Matters for EU Inc and Startups Aleph Alpha's move shows that Europe can build its own large language models without compromising on sovereignty. For anyone following EU Inc news or considering incorporating a startup in Europe, this is a proof point: the ecosystem is maturing. You don't have to rely solely on US or Chinese AI giants. And there's more. Back in April, Aleph Alpha announced a deal with Canadian AI firm Cohere to create a joint transatlantic sovereign AI company. The merger is pending regulatory approval, but the combined entity will operate globally as Cohere with headquarters, R&D centers, and leadership in Canada and Germany. ### The Bottom Line With Kolibri now available and a merger in the pipeline, Aleph Alpha is proving that sovereign AI isn't just a buzzword. It's a real, viable option for governments and enterprises that want control, compliance, and top-tier performance. If you're building a startup in Europe, keep an eye on this space—it could shape the infrastructure you build on.