Research Aleph Alpha On German Reunification Day, the AI company Aleph Alpha released Kolibri, a new open-weight Mixture-of-Experts Transformer model with 78B total parameters and 3B active parameters. The model supports context lengths of up to 1M tokens and is released under the Apache 2.0 license via Hugging Face. Kolibri represents a significant evolution from the company's earlier Kolibri Origin model, a 30B parameter model with a 65k token context window.
The development team leveraged a robust training pipeline to run hundreds of ablation experiments and achieve stable pre-training despite hardware failures. Kolibri is specialized for German language, reasoning, math, and agentic behavior, targeting sovereign mission-critical work in regulated sectors like public administration, industrials, and aerospace. The model optimizes the trade-off between capability and deployment costs, positioning itself on the Pareto frontier for quality versus serving cost in both English and German.
Aleph Alpha emphasizes full supply-chain integrity and transparency, offering customers deployment freedom and intellectual-property safety as inherited compliance properties. The model's small and efficient size allows it to run effectively on-premise, avoiding the need to send internal data to third-party inference services. Read the full tech report for detailed benchmark analysis.
Source: Hacker News · Summarized by HeadlinesBriefing