| Entity | Aleph Alpha |
|---|---|
| Counterparty | Cohere |
| Date | 5 Oct 2026 |
| Region | Europe / DACH |
Aleph Alpha said on 5 October 2026 that it has released Kolibri, a large language model for public administration and industry, developed and trained in Europe and supporting German and English. The model has been available since 3 October for deployment on infrastructure the customer controls. A separate agreement with Cohere to build a transatlantic company remains subject to regulatory approval; until closing, Aleph Alpha operates independently.
Who this affects
- For exporters: The notice states no export licence or transfer restriction. Kolibri is described as deployable on infrastructure the customer controls, with licence terms posted on Hugging Face.
- For investors: The Cohere agreement is still subject to regulatory approval. Until closing, Aleph Alpha continues to operate independently. The company says it employs some 200 people across four locations in Germany, with headquarters in Heidelberg.
- For counsel/compliance: The technical report is cited for copyright, data-protection, and EU AI Act measures. Training data was screened against a blocklist of more than 4.5 million URLs, including the European Commission’s Piracy Watch List, and each third-party dataset was assessed for licence terms, lawful sourcing, and opt-outs. Applicable licence terms are on Hugging Face, not named in the notice.
- For policymakers: The stated use is mission-critical applications in public administration and industry. The notice points to the technical report for the EU AI Act measures taken.
The read
The notice splits the product from the company transaction. Kolibri has been available since 3 October, while the Cohere agreement is still unclosed and Aleph Alpha says it operates independently until closing. The licence is not named here; it is deferred to Hugging Face. The compliance claim is a screening record — a blocklist of more than 4.5 million URLs and per-dataset licence and opt-out checks — not a stated approval.
What was the action?
Aleph Alpha released Kolibri, a specialized large language model for public administration and industry. It supports German and English, was developed and trained in Europe, and has been available since 3 October. It combines 78 billion total parameters with around 3 billion active per token, in a mixture-of-experts architecture. German-language text accounts for around 23% of pre-training data.
What’s the penalty / scope, and how is it set?
No penalty is stated. The action is a model release. Scope is deployment on infrastructure the customer controls, with documented data provenance and design decisions. The model supports agentic workflows, retrieval-augmented generation, and native tool calling, and is trained to abstain when supplied documents do not contain sufficient evidence. Licence terms and documentation are at the Hugging Face repository Aleph-Alpha/Kolibri-1.
Who else is exposed?
Cohere is the only other company named. Aleph Alpha says it recently announced an agreement with Cohere to build a transatlantic sovereign AI company; that transaction remains subject to regulatory approval. Intended users are named only as public administration and industry. No customer is named.
What happens next
The open item on this record is the Cohere transaction, which remains subject to regulatory approval. Until closing, Aleph Alpha continues to operate independently. The notice gives no approval body, filing number, or closing date. Kolibri itself is already available, since 3 October.
All training data was screened against a blocklist of more than 4.5 million URLs, drawing on sources including the European Commission’s Piracy Watch List.Aleph Alpha Release, 2026 →