Aleph Alpha open-sources Kolibri: 78B MoE with 3B active params and 1M context, Apache 2.0
TejasKumar_ · x · 2026-10-04
German sovereign AI company Aleph Alpha released Kolibri, an English-German Mixture-of-Experts transformer with 78B total and 3B active parameters, up to 1M-token context, full weights on Hugging Face under Apache 2.0. Built for self-hosted, regulated-industry workloads, it needs 2x 80GB A100/H100 or one H200/B200/B300. It came out of a hardened training pipeline first validated on Kolibri Origin (30B total/3B active, 65k context), and community commentary suggests pairing it with lightweight most-embed-de embeddings for a fully on-prem help-desk assistant.
More from Models
- Daniel Han publishes summary of LLM benchmarks you can actually trust — danielhanchen · 2026-10-06
- Claim Verification Benchmarks Mostly Test Retrieval, Not Reasoning, Finds 24K-Trace Study — deliprao · 2026-10-06
- COLM26 study: LLMs ace claim verification benchmarks by taking shortcuts, not verifying — deliprao · 2026-10-06
- Opus 5.5 uses 26k tokens vs Astra's 12k yet costs 23% less per task at equal AA score — ChrisGPT · 2026-10-06
- GPT-6 Astra claimed to be first AI crossing world-class astrophysics threshold — johnseach · 2026-10-06
- $500/mo AI subscription is huge money in Jakarta: PPP pricing debate — sujingshen · 2026-10-06