European open-source model Kolibri is here: 3.46B activated parameters, supporting 1M context.

动察 Beating AI News: German AI company Aleph Alpha has released and open-sourced Kolibri, a bilingual English-German model. The model uses a MoE architecture with a total of 78.1 billion parameters, but only about 3.46 billion parameters are activated per token. The weights have been uploaded to Hugging Face under the Apache 2.0 license.

Kolibri supports tool calling and adjustable reasoning modes, and its context can be extended up to about 1 million tokens. However, the official recommendation is to keep daily deployment within 262,000 tokens, which is also the longest context the model was actually trained on.

In Aleph Alpha's official evaluation, Kolibri scored 96.9% on AIME 2025 and 85.9% on LiveCodeBench v6. Its overall English-German results are slightly higher than Qwen3.5 35B-A3B, which also has about 3B activated parameters, but still lower than the fully activated 27B Qwen3.8.