Aleph Alpha releases Kolibri weights for deployment on organizations' own infrastructure

Aleph Alpha released Kolibri on October 3, making the AI model's full weights available on Hugging Face under the Apache 2.0 license, ForkLog reported. It targets government organizations and regulated industries, which can deploy it on their own infrastructure without sending internal data to third-party cloud AI providers. Kolibri supports reasoning, external tool use and a context window of 1,048,576 tokens. Its architecture activates approximately 3.46 billion of its 78.1 billion parameters for each token processed. The FP8 version occupies about 78 GB of memory. Aleph Alpha specifies a minimum deployment configuration of one Nvidia H200, B200 or B300, or two A100 80 GB/H100 SXM5 GPUs. Its reported benchmark results come from company tests and are not independently verified.