GoKawiilAleph Alpha releases Kolibri, a 78B-parameter open-weight LLM for German and English
Aleph Alpha, a German AI company, released Kolibri on 3 October 2026 under the Apache 2.0 license, with weights published on Hugging Face. It is a mixture-of-experts model with 78.1 billion total parameters but only 3.46 billion active per token, trained from scratch on 768 NVIDIA B200 GPUs across data centers in Germany and Finland on roughly 24 trillion tokens, over a fifth of them German.
GoKawiil's interpretation of the reporting above, not reported fact.
The company says Kolibri was designed with the EU AI Act in mind 'from the ground up,' positioning it as a model built to comply with European regulatory requirements rather than retrofitted afterward. According to Aleph Alpha's own evaluation, it outperforms other models of similar size in both German and English, though independent verification of those claims has not been confirmed. Its efficient architecture and open license could make it attractive to European developers seeking regulation-compliant alternatives to larger US or Chinese models.
- Kolibri uses a mixture-of-experts design, activating only 3.46 billion of its 78.1 billion parameters per token
- Released under Apache 2.0 with weights on Hugging Face, trained entirely on infrastructure in Germany and Finland
- Aleph Alpha states the model was built with EU AI Act compliance considered from the start of development
Source: tej.as — Tejas Kumar, 2026-10-03
Published there as: “Kolibri is an open-weight LLM from Aleph Alpha for German and English”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.