- The /teamwork multiagent system solves years-old research problems alongside human experts
- The system is built on the Gemini 4 Argon model
- Google opens the tools externally via a researcher access program
AI CURATED HOT STORY · 11/14/2026
Google's Multi-Agent Teamwork Formalizes the Weil Conjecture
Google Research has introduced Teamwork, a multi-agent system within its Antigravity platform powered by Gemini models, which has formally proved the Weil Conjecture, generating approximately 1.8 million lines of Lean code. The system also reproduced a formal proof of finite-time blow-up for the three-dimensional Navier-Stokes equations and achieved verifiable results in molecular structure recognition and C++ to Rust code rewriting, marking a shift of AI from an auxiliary tool to a research partner.
Teamwork employs an architecture of specialized agents solving problems in parallel, complemented by independent questioners and reviewers to reduce errors in long-chain reasoning. The Weil Conjecture formalization required about 28,800 sub-agent calls, molecular recognition achieved 95% Top-1 and 100% Top-3 accuracy, and the libgav1 decoder was rewritten into approximately 170,000 lines of Rust code. Google is providing 10,000 academic researchers with six months of free high-quota access through Antigravity, and the tools are available via a researcher access program.
This achievement has profound implications for mathematical formalization, AI for Science, and software engineering. The multi-agent collaboration solving long-standing open problems demonstrates AI's potential in complex research tasks, promising to accelerate scientific discovery and engineering practice, and driving a paradigm shift of AI from auxiliary tool to research partner.
Hourly heat curve
46 fully observed hoursThe line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.
Coverage timeline
Total 3 reports · Latest first- Hacker News AIT2·78 pts
- IT之家 · 智能时代T2·88 pts
- Teamwork pairs parallel specialist agents with independent critic and reviewer agents to curb long-horizon reasoning errors
- Formalizing the Weil conjectures took about 28,800 sub-agent calls and roughly 1.8 million lines of Lean code
- Molecular identification reached 95% Top-1 and 100% Top-3 accuracy, while libgav1 was rewritten in about 170,000 lines of Rust
- Google offers 10,000 academic researchers six months of free high-quota access via Antigravity
- Google ResearchT1·78 pts
- The /teamwork multiagent system collaborates with human experts to solve long-open problems in math, chemistry, systems software, and ML engineering
- Google opens the tools externally via a researcher access program to drive adoption
- The work ties into Gemini 4 Argon and positions AI as a research partner