• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
Technology
Report

Aleph Alpha releases Kolibri-1 with open weights

An October 4 tech roundup says the German firm’s 78B model carries an Apache 2.0 license.

IV
AI
S|
3 Sources, 20h ago, first seen 20h ago

TLDR

An October 4 tech roundup says Aleph Alpha shipped Kolibri-1, a 78B open-weight model with an Apache 2.0 license. It also says the model hit the top of Hacker News.

Combined views

1.1K

3 Sources, first seen 20h ago

32 likes9 comments3 saves4 reposts

Combined views

1.1K

3 Sources, first seen 20h ago

32 likes9 comments3 saves4 reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Featured Source

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

3 Sources

@ivke2006🚀 Daily Top 10: AI • ICT • IT • Robotics – October 4, 2026 1️⃣ The guy who wrote OpenAI’s safety reports just quit. David Robinson, 3.5 years in: the culture is broken. 2️⃣ Amazon is throwing $1B+ at the towns hosting its data centers, and dropping the NDAs. The backlash did not wait for the check. 3️⃣ Germany’s Aleph Alpha shipped Kolibri-1. Open weights, Apache 2.0, 78B. It hit the top of Hacker News. 4️⃣ FieldAI raised $700M at a $10B valuation for a “universal” robot brain. Humanoids, dogs, drones. Revenue already past $135M. 5️⃣ Claude is moving into healthcare. Billing, prior auth, and a link to Apple Health and Android Health Connect. 6️⃣ Broadcom is syndicating ~$60B in debt so Anthropic can keep buying chips. Anthropic is still aiming at a mid-November IPO. 7️⃣ ChatGPT gets virtual try-on. Upload a photo, see the clothes. Shopping agents just got a mirror. 8️⃣ Altman says he is “very uncomfortable” with people treating models like a religion. He called it a real safety issue. 9️⃣ Frontier models beat licensed CPAs on structured accounting tasks. Narrow work. Still a bad week for the month-end close. 🔟 Meta open-sourced Muse Gadgets. A Raspberry Pi can now be a screen or button for its agent. Which one do you like? Let us know!👇20h
@AistmaiGood morning, folks. If an agent already touches your orders, your tickets or your bookings, start with the first line below. 🔹 A new agent benchmark grades the database, and the best model got all 20 runs right on 47.53% of tasks — Hugging Face. 🔹 Kolibri, an open-weight English-German model with a 1M-token context, is out under Apache 2.0 — Aleph Alpha. 🔹 OpenAI's safety reports lead quit, saying its culture is broken, after OpenAI warned 100+ organisations of rogue agents — The Guardian. 🔹 Amazon says it no longer uses NDAs with government agencies, with over 100 data center moratoriums pending — TechCrunch. 🔹 A crop of AI assistants now lives in iMessage and SMS instead of an app, with Instinct valued at $10 billion — TechCrunch. The move today: take one task your agent already runs, repeat it from a clean start, and grade the records it leaves behind. Follow @Aistmai for the morning brief and more posts like this.16h
@SKatalystAI🇩🇪 Germany just dropped a new open-weight AI model! Aleph Alpha released Kolibri-1 yesterday: >78B total / ~3.5B active MoE >1M context >German + English >Reasoning + tool calling >Apache 2.0 open weights and some of the claimed benchmarks are surprisingly strong: >AIME 2025: 96.9% >GPQA Diamond: 84.3% >LiveCodeBench v6: 85.9% >SWE-Bench Verified: 66.4% What interests me more is the design choice: only ~3.5B parameters active per token while targeting sovereign/on-prem deployments for European industry and government. One big caveat: these are Aleph Alpha's own evaluations The weights are already available, so now I want to see how Kolibri behaves on an independent repo!10h
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    3 Sources

    @ivke2006🚀 Daily Top 10: AI • ICT • IT • Robotics – October 4, 2026 1️⃣ The guy who wrote OpenAI’s safety reports just quit. David Robinson, 3.5 years in: the culture is broken. 2️⃣ Amazon is throwing $1B+ at the towns hosting its data centers, and dropping the NDAs. The backlash did not wait for the check. 3️⃣ Germany’s Aleph Alpha shipped Kolibri-1. Open weights, Apache 2.0, 78B. It hit the top of Hacker News. 4️⃣ FieldAI raised $700M at a $10B valuation for a “universal” robot brain. Humanoids, dogs, drones. Revenue already past $135M. 5️⃣ Claude is moving into healthcare. Billing, prior auth, and a link to Apple Health and Android Health Connect. 6️⃣ Broadcom is syndicating ~$60B in debt so Anthropic can keep buying chips. Anthropic is still aiming at a mid-November IPO. 7️⃣ ChatGPT gets virtual try-on. Upload a photo, see the clothes. Shopping agents just got a mirror. 8️⃣ Altman says he is “very uncomfortable” with people treating models like a religion. He called it a real safety issue. 9️⃣ Frontier models beat licensed CPAs on structured accounting tasks. Narrow work. Still a bad week for the month-end close. 🔟 Meta open-sourced Muse Gadgets. A Raspberry Pi can now be a screen or button for its agent. Which one do you like? Let us know!👇20h
    @AistmaiGood morning, folks. If an agent already touches your orders, your tickets or your bookings, start with the first line below. 🔹 A new agent benchmark grades the database, and the best model got all 20 runs right on 47.53% of tasks — Hugging Face. 🔹 Kolibri, an open-weight English-German model with a 1M-token context, is out under Apache 2.0 — Aleph Alpha. 🔹 OpenAI's safety reports lead quit, saying its culture is broken, after OpenAI warned 100+ organisations of rogue agents — The Guardian. 🔹 Amazon says it no longer uses NDAs with government agencies, with over 100 data center moratoriums pending — TechCrunch. 🔹 A crop of AI assistants now lives in iMessage and SMS instead of an app, with Instinct valued at $10 billion — TechCrunch. The move today: take one task your agent already runs, repeat it from a clean start, and grade the records it leaves behind. Follow @Aistmai for the morning brief and more posts like this.16h
    @SKatalystAI🇩🇪 Germany just dropped a new open-weight AI model! Aleph Alpha released Kolibri-1 yesterday: >78B total / ~3.5B active MoE >1M context >German + English >Reasoning + tool calling >Apache 2.0 open weights and some of the claimed benchmarks are surprisingly strong: >AIME 2025: 96.9% >GPQA Diamond: 84.3% >LiveCodeBench v6: 85.9% >SWE-Bench Verified: 66.4% What interests me more is the design choice: only ~3.5B parameters active per token while targeting sovereign/on-prem deployments for European industry and government. One big caveat: these are Aleph Alpha's own evaluations The weights are already available, so now I want to see how Kolibri behaves on an independent repo!10h
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet