MENU

Kolibri 1 explained: Aleph Alpha’s open-weight MoE LLM specs

ActuIA Germany
Overview
German AI firm Aleph Alpha launched ‘Kolibri 1,’ an open-weight Mixture-of-Experts (MoE) large language model, under the Apache 2.0 license on October 3, 2026. This 78.1 billion parameter model activates only 3.46 billion parameters per token, achieving high efficiency and superior performance over models like Nemotron 3 Super and GPT-OSS 120B in both English (75.5) and German (70.8) benchmarks. The open-source release is a significant step towards fostering community-driven AI innovation and establishing European technological autonomy in the global AI landscape.
In Depth

Key Findings

On October 3, 2026, marking German Unity Day, Aleph Alpha, a prominent German AI startup, unveiled its open-weight Mixture-of-Experts (MoE) large language model, ‘Kolibri 1.’ This groundbreaking model features 78.1 billion total parameters but efficiently activates only 3.46 billion parameters per token, balancing vast scale with computational efficiency. Crucially, its weights are available for download on Hugging Face under the permissive Apache 2.0 license, promoting transparency and broad community engagement.

Technical Details

  • Model Name: Kolibri 1
  • Architecture: Mixture-of-Experts (MoE) large language model
  • Total Parameters: 78.1 billion
  • Active Parameters per Token: 3.46 billion
  • License: Apache 2.0 (open-weight)
  • Distribution: Weights downloadable from Hugging Face
  • Language Specialization: Optimized for German and English
  • Benchmark Scores:
    • English: 75.5 (outperforming Nemotron 3 Super and GPT-OSS 120B in specific tasks)
    • German: 70.8
  • Accompanying Resources: A technical blog post, a 189-page technical report, and a public summary of training data are provided.

The adoption of the MoE architecture allows Kolibri 1 to activate only the necessary parts of its vast parameter space for a given task, significantly boosting computational efficiency. This design enables the model to achieve high performance on complex tasks while minimizing resource consumption. Its strong benchmark performance in both German and English holds particular significance for multilingual AI development.

Background & Context

The release of open-weight models is vital for fostering collaboration and innovation within the AI research community. The Apache 2.0 license, which permits commercial use, modification, distribution, and sublicensing while granting patent rights, provides a robust framework for companies and researchers to confidently adopt and build upon the model. This move by a European firm like Aleph Alpha to release a high-performing open model is strategically important in a global AI market largely dominated by major U.S. and Chinese tech giants, contributing to Europe’s technological sovereignty and competitive edge.

Strategic Significance & Outlook

The open-weight release of Kolibri 1 is expected to accelerate the adoption of AI technology in German-speaking regions and catalyze broader community-driven innovation. Developers and enterprises will find it easier to leverage this model as a foundation for building customized AI solutions tailored to specific industries and applications. Combining high performance with an open license, Kolibri 1 is poised to become a powerful tool for researchers, startups, and large corporations alike, contributing significantly to the vibrancy and growth of the global AI ecosystem.

Source: https://www.actuia.com/en/news/aleph-alphas-kolibri-what-the-apache-20-license-guarantees/

Get our weekly technology intelligence — free

Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.

Subscribe Free — Weekly Tech Intelligence

By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.

  • Your email and selected fields are used only to deliver the newsletter.
  • We never share your information with third parties.
  • You can unsubscribe anytime via the link in each email.

See our Privacy Policy for details.

Takes about a minute · Unsubscribe anytime

Published by Troy-Technical, an independent site run by one engineer with a career in materials development.
About the author / Contact info@troy-technical.jp
Let's share this post !

Author of this article

TOC