Google Antigravity and Gemini 3.7 Flash Solve Math and Engineering Tasks

Google has introduced substantial updates to its Antigravity research platform, integrating the Teamwork framework with the Gemini 3.7 Flash model to demonstrate enhanced capabilities in autonomous multi-agent problem-solving. Teamwork operates as an orchestration system enabling AI agent teams to collaborate, critique, and iterate over extended periods, effectively addressing complex, long-horizon challenges across academic and engineering fields. In theoretical computer science and mathematics, the system resolved seven open problems accepted at top-tier venues such as FOCS and JMLR. Key accomplishments include verifying Knuth's Cycles Conjecture via a Lean proof spanning more than 40 pages, as well as advancing solutions for sparse convex optimization, provable LLM quantization, and prefix-matrix factorizations. The framework achieved a 71% score on the TCSBench benchmark. For systems engineering, the model produced a cycle-accurate, out-of-order RISC-V CPU simulator developed from scratch. This implementation successfully boots the xv6 operating system to a functional shell, maintaining a cycle alignment error of only 0.71% compared to hardware ground truth. In open-source software development, the initiative yielded direct contributions to core libraries. The team integrated performance optimizations into Eigen by adding SIMD fast-paths and improved ParlayHash performance by doubling insert throughput while reducing memory consumption by 25%. These outcomes illustrate the effectiveness of coupling advanced language models with structured multi-agent workflows to drive measurable progress in both theoretical research and software infrastructure.