Aleph-Alpha/Kolibri-1 — new model trending #28 on Hugging Face
Aleph Alpha released Kolibri-1, an Apache 2.0 78B MoE reasoning model for German and English.
Aleph Alpha released Kolibri-1, an Apache 2.0 mixture-of-experts reasoning model aimed at German and English, with tool calling and an explicit reasoning mode. It has 78,103,074,560 total parameters and 3,457,573,120 active per token, using 384 experts per layer with one shared and six routed. Weights are float8 in 128 by 128 blocks; the footprint is about 78 GB. The model was pre-trained on 20 trillion tokens on 768 NVIDIA B200 GPUs for 21 days, followed by 3.44 trillion mid-training tokens and a long-context stage, for about 6.4e23 FLOPS. Native context is 262,144 tokens, with quality checked up to 1,048,576; the knowledge cutoff is 18 June 2026.
72