Aleph Alpha's Kolibri Lags Behind Open-Weight Leaders
Aleph Alpha has released Kolibri, a new open-weight German-English AI model under the Apache-2.0 license, featuring a Mixture-of-Experts architecture with 78 billion total parameters and about 3.46 billion active per token. Trained on 20 trillion tokens, with 21.3% in German, Kolibri supports context lengths up to one million tokens and is designed for regulated and industrial applications, emphasizing data sovereignty for German public sector and industry clients. While it performs well against earlier models like Qwen3.6 and Mistral Small 4, independent benchmarks rank it behind at least 20 other open-weight models. The company is merging with Cohere, and its brand will disappear. In a broader economic debate, Reid Hoffman argues AI data center expansion is the 'only reason' the US avoids recession, though economists dispute this, adding complexity to discussions on AI's economic impact.
- •Aleph Alpha released Kolibri under Apache-2.0 license, with 78 billion total parameters and 3.46 billion active per token.
- •Kolibri was trained on 20 trillion tokens, 21.3% in German, and supports context lengths up to one million tokens.
- •Independent benchmarks place Kolibri behind at least 20 other open-weight models.