Aleph Alpha releases Kolibri-1, a 78B open-weight reasoning model

Aleph Alpha has released Kolibri-1, a 78-billion-parameter open-weight reasoning model under the Apache-2.0 license. The model uses a mixture-of-experts architecture that activates 3.46 billion parameters per token. It supports both German and English, includes tool-calling capabilities, and can handle context windows of up to one million tokens.

The release expands sovereign deployment options for organizations that require long-document processing or local control over their AI infrastructure. Full model operation still carries substantial memory requirements.

Source: https://huggingface.co/