Model
Kolibri 1
A Mixture-of-Experts reasoning large language model released by Aleph Alpha for German and English. It uses sigmoid routing and an architecture that alternates sliding-window attention with full attention.
Timeline (last 90 days)
Oct 15
Aleph Alpha's Kolibri 1 is a MoE reasoning model focused on German and English, in which each of the 50 layers routes each token to 6 of 384 experts via sigmoid routing, plus an additional shared expert.
Unverified50%