Kolibri-1

Kolibri-1 is Aleph Alpha's sovereign open-weight text LLM, released on 2026-10-03 under the Apache-2.0 license with weights on Hugging Face (an FP8 flagship "Kolibri-1" plus a BF16 base "Kolibri-1-BF16" and several quantized variants). It is a Mixture-of-Experts model with 78.1B total (78,103,074,560) / ~3.46B active (3,457,573,120) parameters — 50 layers, 384 routed experts, 6 active per token — tuned for bilingual German/English use and trained natively to a 262,144-token context, extensible to 1,048,576 tokens (1M). Text in / text out, with training for grounding, tool-calling and abstention when evidence is insufficient; built and trained in Germany and Finland. Aleph Alpha reports post-training scores of 75.5% overall (English) and 70.8% (German), 96.5% math and 89.3% code (English averages), and 63.2% on RULER at 1M-token context. Knowledge cutoff 2026-06-18; figures are vendor / self-reported.