Aleph Alpha releases Kolibri, an open German and English AI model built for EU rules

German AI company Aleph Alpha has released Kolibri, an open-weight model for German and English, under the Apache 2.0 licence on Hugging Face. It has 78.1 billion parameters but uses only 3.46 billion for each token it produces, which keeps it cheaper to run, and it can read up to about a million tokens at once. Training ran in Germany and Finland, and the design targets the EU AI Act, GDPR and the EU code of practice for general-purpose AI, with personal data removed before training.

• If you serve German-speaking customers or handle sensitive documents, ask your software supplier whether an EU-hosted Kolibri option could cover work you would rather not send to US or Chinese AI services.
• Do not plan to run it on an office PC: it needs a data-centre GPU such as one H200 or two H100s, so it will reach you through hosted services, not as a laptop app.

Source: MarkTechPost. This is our short summary of their reporting. The paragraphs marked "Our read" and the things to do are SeedFoundry's opinion. The summary is drafted with AI tools. Image: Aleph Alpha, from its own announcement. How we source and credit