Aleph Alpha releases open-weight model Kolibri 1
What happened
On October 3, 2026, Hacker News' front page reported that Aleph Alpha released Kolibri, a bilingual English-German open-weight MoE model. Early reports said it has 78B total parameters and about 3B active parameters, with context up to 1M tokens. Later reports gave the name as Kolibri 1, with 78.1 billion total parameters and 3.46 billion activated per token, distinguishing a native context of 262,144 tokens from a tested extended length of 1,048,576 tokens. The later reports also state the model weights and config files use the Apache 2.0 license.
Written by AI from the coverage · updated 1 hour ago
Coverage
Follow the reports to see the story from different sides.
- Hacker News front pageShow HN: Germany's new sovereign AI model Kolibri
Aleph Alpha 于 2026 年 10 月 3 日发布德英双语开放权重模型 Kolibri 1，权重和配置文件采用 Apache 2.0 许可。模型采用 MoE 架构，总参数为 78.1 billion，每个 token 激活 3.46 billion，原生上下文为 262,144 tokens，测试扩展至 1,048,576 tokens。
- Hacker News front pageKolibri Has Landed: A Sovereign Open-Weight Model
Aleph Alpha 发布英德双语 MoE 模型 Kolibri，拥有 78B 总参数、约 3B 激活参数，支持最长 1M tokens 上下文。
Heat over time
Not enough continuous observations to draw a trend yet.