Kolibri is an open-weight language model from Aleph Alpha for English and German, built for regulated work such as public administration and aerospace.
In plain words
Kolibri is a mixture-of-experts model: it has 78 billion parameters in total, but only 3 billion do the work for each token, which keeps it cheap to run. Aleph Alpha trained it for German as well as English and puts the complete weights on Hugging Face under the Apache 2.0 licence. It can read up to 262,144 tokens at once. Aleph Alpha says that on maths, coding and long-context tasks it does as well as models using as many as four times the active parameters.
A quick example
Imagine a city office that wants an assistant to draft replies to residents' letters in German. It runs Kolibri on its own servers, and a staff member checks each draft before it is sent.
Where you'll meet it
- German and English assistants and agent setups where a person checks the output.
- Regulated work such as public administration, industry and aerospace.
Related
Sources
- DocsKolibri Has Landed: A Sovereign Open-Weight Model, Aleph Alpha · read 3 Oct 2026
- DocsAleph-Alpha/Kolibri-1, Aleph Alpha (Hugging Face) · read 3 Oct 2026
Checked by WhatIsAI against the sources above · 3 Oct 2026. A full five-minute explainer with a diagram may follow. Spotted an error? Report it.