This weekly update highlights major breakthroughs in artificial intelligence (AI) agents and scientific discovery. A team of researchers created an AI scientist called Kosmos that completed six months of human research work in just twelve hours, achieving 79.4% accuracy. The AI system read 1,500 scientific papers and wrote 42,000 lines of computer code to make new discoveries. Meanwhile, scientists discovered that multiple AI agents working together perform better than single AI models, similar to how team members solve problems better than individuals. Several new powerful AI models launched this week, including Baidu's ERNIE model from China, which outperformed competitors like GPT and Gemini on visual tests. Microsoft introduced two new AI models called MAI-Voice-1 and MAI-1-preview as part of its independent AI development. Moonshot AI from Beijing released Kimi K2 Thinking, an open-source model costing only $4.6 million to build. Google added Gemini's Deep Research feature to NotebookLM to help users create detailed research reports. OpenAI launched GPT-5.1 with improved personality and reasoning abilities. Researcher Fei-Fei Li explained that AI progress depends on world models that understand physical space and how environments change. The biggest trend in 2025 is that specialized AI agents working as teams are becoming more important than single general-purpose AI tools.