Microsoft Just Dropped KOSMOS: AI With 80% Human-Level Performance

Microsoft just dropped KOSMOS — an autonomous AI system that runs for 12 hours straight, spins up hundreds of smaller AIs in sync, reads over 1,500 papers, writes 40,000 lines of Python, and delivers reports with ~80 % accuracy in early reviews.
At the same time, Google’s DS-STAR turns chaotic business data into working analysis by planning, coding, testing, and auto-debugging its own Python. And Moonshot AI’s Kimi K2 Thinking pushes open-source reasoning to new limits with hundreds of chained tool calls for browsing, math, and coding. The global AI race just hit another level.

???? Brand Deals & Partnerships: me@faiz.mov
✉ General Inquiries: airevolutionofficial@gmail.com

What You’ll See:
0:00 Intro
0:32 KOSMOS — An AI Scientist for Autonomous Discovery
https://arxiv.org/abs/2511.02824
4:09 Mustafa Suleyman – Towards Humanist Superintelligence (Microsoft AI blog)
https://blogs.microsoft.com/blog/2025/11/02/towards-humanist-superintelligence/
5:43 Moonshot AI — Kimi K2 Thinking (Open-source reasoning model; 200–300 tool calls; test-time scaling)
https://kimi.moonshot.cn/
Reuters — China’s Moonshot AI launches open-source Kimi K2 Thinking model (industry coverage)
https://www.reuters.com/technology/chinas-moonshot-ai-launches-open-source-kimi-k2-thinking-model-2025-10-29/
8:16 Google Research Blog — DS-STAR: A state-of-the-art versatile data science agent
https://research.google/blog/ds-star-a-state-of-the-art-versatile-data-science-agent/