Aleph Alpha releases Kolibri, an open-weight model that makes the case for European AI sovereignty
decoder ·
Aleph Alpha has released Kolibri, a new open-weight language model designed to promote European AI sovereignty. The German-English model features 78 billion parameters with approximately three billion active per token using a mixture-of-experts architecture. It claims to sit on the Pareto front of quality and operating cost, outperforming comparable models from March and April 2026. Training data includes a dedicated German pipeline comprising 21.3 percent of the dataset, supplemented by synthetic data generated by Chinese models. Kolibri achieves a 71% score on German benchmarks while decoding faster than competitors like GPT OSS A5B and Qwen 3.6 A3B. The model supports context windows of up to one million tokens and was trained on 768 B200 GPUs in Germany and Finland. Developed with the EU AI Act in mind, it targets sectors such as public administration, aviation, and industry. Weights are available under an Apache 2.0 license on Hugging Face.
decoder