Google's Cogentic uses multi-agent Gemini system to find math proofs
omarsar0 · x · 2026-10-03
Google Research's Cogentic is a multi-agent proof-discovery system running on Gemini that tackles open problems in theoretical CS from the problem statement alone, with no expert hints. It works in rounds: an orchestrator decides how many provers to run, each prover gets one direction (a bound or counterexample search) plus a briefing written independently by summarizer agents from prior attempts and verifier feedback. The pattern — loose execution structure plus dedicated advisor and verifier agents — is argued to generalize broadly.
Related event: Google's Cogentic Multi-Agent System Tackles Open Math Problems(4 posts)→
More from Research
- Google team shows neural cellular automata can solve mazes, Sudoku and ARC-AGI-1 — burny_tech · 2026-10-03
- Constant-Size Memory for Video World Models Proposed in New arXiv Paper — plsendfast · 2026-10-03
- NUS, Tsinghua, PKU and 4 more universities open-source OpenWAM world-action foundation model — jiqizhixin · 2026-10-03
- Pedagogical RL: using privileged info to actively sample rollouts instead of blind scoring — lateinteraction · 2026-10-03
- Amazon & Duke study: supervising just 1 token per rollout boosts LLM reasoning — 机器之心 · 2026-10-03
- RNE: an open-source Rust robot-native engine for deterministic sim and policy evaluation — rsasaki0109 · 2026-10-03