Google Combines Antigravity and Gemini 3.7 Flash to Solve Multi-Agent Math and Engineering Problems

Google enhanced multi-agent collaboration capabilities for solving long-horizon complex problems by combining Gemini 3.7 Flash with Antigravity's Teamwork framework. This combination accelerated problem-solving across research and engineering.

Key Achievements

• Mathematics and Theoretical Computer Science: Solved 7 open problems from major journals such as FOCS and JMLR. Verified Knuth's Cycles Conjecture in Lean (with a proof exceeding 40 pages), including sparse convex optimization, LLM quantization proofs, and prefix matrix decomposition. Achieved a score of 71% on TCSBench.
• Systems Engineering: Built an out-of-order RISC-V CPU simulator with cycle-accurate precision from scratch. Booted the xv6 operating system up to the shell, achieving a cycle alignment error of 0.71% compared to hardware benchmarks.
• Open Source Software: Upstreamed performance optimization code to core libraries such as Eigen (SIMD fast paths) and ParlayHash (2x increase in insertion throughput, 25% reduction in memory).