Back to Model
K

Kolibri 1

Model Oct 5, 2026

Aleph Alpha's German-English open-weight reasoning model for self-hosted deployments.

Visit primary link 2 sources attached

What happened lately

Kolibri 1 timeline

Kolibri 1 at a glance

Aleph Alpha released Kolibri 1 on October 3, 2026. The German-English mixture-of-experts model has 78.1 billion total parameters and activates about 3.46 billion per token. Its weights and configuration are available on Hugging Face under Apache 2.0.

Deployment and limits

Aleph Alpha says Kolibri supports reasoning modes, tool calls and contexts of up to one million tokens. Its model card says training reached 262,144 tokens and recommends that shorter limit for complex or latency-sensitive use. Serving requires Aleph Alpha’s inference package and substantial GPU memory. The Apache 2.0 grant covers the published weights and configuration files, not the company’s other code or training methods. Benchmark comparisons in the release are provider-run and should be tested on the intended workload.