The German AI research lab designed its latest model to adhere to the General-Purpose AI Code of Practice and the GDPR from its inception. This regulatory alignment includes a specific focus on copyright law to develop what the company calls trustworthy technology. To enhance regional utility, Aleph Alpha utilized a bilingual German-English tokenizer and ensured organic German content comprised 21.3% of the system's pre-training tokens.
While the release of Kolibri has been hailed as a win for sovereign European AI, researcher Ethan Mollick noted the model was fine-tuned on synthetic data from Chinese models GLM and Qwen. The system uses a mixture-of-experts architecture with 78B total parameters, 3.46B of which are active, and supports a 1M token context window. It is distributed under an Apache 2.0 license, allowing for local hosting and private infrastructure use.
Content history (1)
- 2026-10-04 · Summary · vi · Wording fixPhòng thí nghiệm nghiên cứu AI Đức đã thiết kế mô hình mới nhất tuân thủ Bộ quy tắc Thực …Phòng thí nghiệm nghiên cứu AI Đức đã thiết kế mô hình mới nhất tuân thủ Bộ quy tắc Thực …
Key sources
- SOURCE@aleph__alpha“78B parameters. 3.46B active. Up to 1M tokens of context. Built in Europe.”x.com
- SUPPORT@testingcatalog“focused on including organic German data throughout the training process of the model, so that 21.3% of the pre-training tokens are German”x.com
- SUPPORT@emollick“Europe's new “sovereign model” is fine-tuned on data generated by… GLM and Qwen”x.com
- SUPPORT@clementdelangue“Small bird, fast wings, Kolibri is here.”x.com
- SUPPORT@aidangomez“Congrats to the AlephAlpha team on a very cool German-English model”x.com
- SUPPORT@cohere“New Apache 2.0 model from @Aleph__Alpha. It's really good.”x.com
- SOURCEhuggingnewshuggingnews.com