Stealthy German lab Aleph Alpha drops open-weight Kolibri: 78B params, 3.46B active, 1M context
skdh · x · 2026-10-05
Aleph Alpha, a relatively unknown German AI lab, has released the open weights of Kolibri, trained from scratch in Europe.
- 78B total parameters with only 3.46B active (sparse architecture)
- Up to 1M tokens of context
- Beats Qwen, Mistral, and Nvidia Nematron models with similar active parameter counts
- Released under Apache 2.0, fully self-hostable
The sudden arrival of a SOTA open-weight model from a "random German lab" is drawing wide attention.
More from Models
- GPT-6 works for 37 hours straight on a single task, wowing developers — tom_doerr · 2026-10-05
- Fan-Made Timeline Maps All 44 Major Qwen Releases, From 7B to 2.4T Open Weights — ai-lover · 2026-10-05
- GPT Astra 6 (Ultra) Called 'Trash' on Codex: Ignored Instructions, 'Chose Convenience Over Scientific Correctness' — raskingballs · 2026-10-05
- Ethan Mollick slams OpenAI's GPTs shutdown: enterprises and educators left behind — emollick · 2026-10-05
- Midjourney CEO shares model "inner world" visualizations: Mistral imagines cookie heists, OpenAI draws ASCII islands — DavidSHolz · 2026-10-05
- FrontierMath Tier 3, predicted by Terence Tao to resist AI for years, now saturated — Eliv_nurotic · 2026-10-05