Overview
Kolibri is a 78B-parameter mixture-of-experts (MoE) large language model with 3B active parameters per token, released by aleph-alpha as an open-weight bilingual German-English model. It features 1M token context, Apache 2.0 licensing on Hugging Face, and sovereign AI positioning for regulated European enterprise and government deployments.
Recent Developments
- 2026-10-03: Official release as Aleph Alpha’s sovereign open-weight LLM for German-English enterprise and government use cases (2026-10-03-aleph-alpha-kolibri-sovereign-llm)
Independent third-party benchmarks were not widely available at release — performance claims should be verified against official specs until independent evaluations publish.