← All model drops
Live drop
Kolibri-1
Aleph AlphaDropped Oct 3, 2026
78B German-English MoE with 3.46B active parameters, 1M context, and explicit reasoning modes for sovereign deployments.
About this release
Live drop
Aleph AlphaDropped Oct 3, 2026
78B German-English MoE with 3.46B active parameters, 1M context, and explicit reasoning modes for sovereign deployments.
About this release
Kolibri-1 is an Apache 2.0 open-weight mixture-of-experts model trained from scratch by Aleph Alpha on 20T bilingual tokens. It activates only 3.46B parameters per token, supports reasoning effort levels and tool calling, and validates up to 1M-token context. FP8 weights (~78 GB) run on a single H200/B200 or two H100 cards via vLLM with custom parsers.
Read the official Aleph Alpha launch pageFirst builds incoming — this lane fills in automatically as people start building with Kolibri-1. Know one? Submit the URL.