Google's Gemini finds new proofs for 5 unsolved math problems

arxiv: arXiv is an online repository for scientific preprints across disciplines including computer science and mathematics. The paper titled 'Cogentic: Multi-Agent Orchestration for Automated Proof Discovery' was posted there with identifier 2609.40324. Gemini: Gemini is Google's large language model used for advanced reasoning tasks. Here, multiple Gemini instances powered the Cogentic system to generate and verify new proofs for previously unsolved math problems. Google: Google is a major technology company that develops and deploys advanced AI systems, including the Gemini family of models. In this development, Google researchers introduced Cogentic, a multi-agent framework built around Gemini to tackle complex mathematical proofs through structured collaboration and verification. Cogentic: Cogentic is a multi-agent orchestration system designed for automated proof discovery in mathematics. It structures Gemini agents to generate ideas in parallel, apply rigorous checking that assumes each step is incorrect until proven, and maintain a shared record of verified components, as detailed in the new arXiv paper. Paper Availability: Details of the Cogentic framework appear in the preprint 'Cogentic: Multi-Agent Orchestration for Automated Proof Discovery' hosted on arXiv. AI Research Approach: Google's Cogentic organizes AI agents like a research team with parallel exploration, strict verification, and persistent memory of proven results rather than relying on single prompts. Mathematics Applications: The Cogentic system enabled Gemini to produce new proofs for five previously unsolved math problems, with human experts confirming each result.