Aleph Alpha released Kolibri on October 5, 2026, saying it targets public administration, aviation, and industry. It is the company's first open-weight model update since its earlier releases.
Aleph Alpha reported 71% on German benchmarks, measured on standard German language tests. That compares with scores from older models released in March and April 2026.
Kolibri is built on a mixture-of-experts architecture and targets public administration, aviation, and industry. Availability begins with open weights on Hugging Face, initially for developers and researchers.
"Kolibri sits on the Pareto front of quality and operating cost in both languages," said Matthias Bastian, the article's author. The model outperforms comparable models like GPT OSS A5B, Qwen 3.6 A3B, and Gemma 4 A4B.
The announcement follows the company's focus on European AI sovereignty, as outlined in the EU AI Act. Aleph Alpha said the model supports context windows of up to one million tokens.
Aleph Alpha did not say how the model will be used in real-world applications, and raised the open question of how it will scale in different industries. The company said it will continue to develop models under European law.
Source: thedecoder
