Aleph Alpha releases Kolibri, an open-weight model that makes the case for European AI sovereignty
Aleph Alpha ships Kolibri: 78B open-weight MoE trained on German silicon, 21% German data. The European sovereignty play is real—but can it compete?
Why it matters
Aleph Alpha releases a German-English open-weight model trained entirely in Europe (768 B200 GPUs, Germany/Finland), positioning it as a sovereignty alternative. The 78B parameter MoE with ~3B active tokens and Apache 2.0 licensing signals commitment to European AI independence, but competitiveness against frontier labs and adoption velocity remain unproven.
The key facts
13 to knowKolibri: 78 billion parameters, ~3 billion active per token (MoE architecture)
Training compute: 768 B200 GPUs in Germany and Finland
Multilingual: 21% of training data is German; German-English focus
License: Apache 2.0 (freely available weights)
Released October 2026
No benchmark scores, latency, or comparative performance data disclosed
No adoption or deployment data provided
European data sovereignty framing (training in EU jurisdiction)
Kolibri: 78B parameters, ~3B active per token (MoE)
21% of training data is German; bilingual German-English focus
Trained on 768 B200 GPUs in Germany and Finland
Weights released under Apache 2.0 license (open-weight)
Aleph Alpha, German startup, positioning on sovereignty narrative
The story so far
Earlier coverage of this storyline
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Aleph Alpha has released Kolibri, a German-English mixture-of-experts model with 78 billion parameters, about three billion of which are active per token. Over 21 percent of the training data is German. The model was trained on 768 B200 GPUs in Germany and Finland, and the weights are freely…