Google's Cogentic uses multi-agent orchestration to produce novel results on five open math problems
KyeGomezB · x · 2026-10-02
Google Research introduces Cogentic, a multi-agent harness for automated proof discovery on open research problems.
Key design:
- An iterative prove-verify loop where an orchestrator allocates independent provers across distinct proof directions;
- Adversarial verification by specialized components, with confirmed intermediate results promoted into a persistent verified ledger that later rounds build on;
- Aimed at research-level math and theoretical CS problems, addressing the limits of single-shot generation for problems requiring exploration of competing conjectures and long-horizon progress.
Using Gemini as the base model, Cogentic produced novel, expert-verified results on five open problems across online learning, auction theory, and mechanism design, each developed in companion papers.
More from coding & agent
- George Hotz seeks a university to host his hands-on ML compilers course; UIUC student says he can make it happen — ctjlewis · 2026-10-02
- Notion ships official agent setup guide positioning it as durable shared memory for AI agents — nbaschez · 2026-10-02
- Non-coder builds full 3D world in 12 hours using 93% of a $20 Claude weekly quota — FinanceYF5 · 2026-10-02
- Coding Agent Index: Sonnet 5.5 tops at 68 but costs $14.19/task vs GPT-6.1 Sol's $1.04 — ArtificialAnlys · 2026-10-02
- Stitch agents now improve this developer's products around the clock — tokumin · 2026-10-02
- Pi 1.0 released with MCP support enabled by default — psychohistorian8 · 2026-10-02