Pairing Google Antigravity with Gemini 3.7 Flash Solves Notable Multi-Agent Math and Engineering Problems

Pairing Google Antigravity with Gemini 3.7 Flash solves notable multi-agent math and engineering problems. On August 31, 2026, Google said it recently launched updates to Teamwork, a framework in Google Antigravity that lets autonomous teams of AI agents collaborate, critique, and iterate over hours or days on complex, long-horizon challenges. Pairing Gemini 3.7 Flash with this multi-agent orchestration accelerated problem solving across research and engineering.

On the research side, Teamwork's Long Proof pattern solved seven open problems across top venues such as FOCS and JMLR. The list includes Knuth's Cycles Conjecture, verified in Lean with 40-plus-page proofs, plus sparse convex optimization, provable LLM quantization, prefix-matrix factorizations, coresets for Lp subspace approximation, and maximal inner product embeddings. Five of the resulting papers are on arXiv. Human experts reviewed the results as correct, except Knuth's Cycles, which was formally verified. Google also reports 71% on TCSBench, its internal theoretical computer science suite, using Gemini 3.7 Flash together with Gemini 3.1 Pro — the highest score in its internal testing. Three of the seven problems were reproduced with Gemini 3.7 Flash, which Google calls the first time a Flash-tier model produced this kind of PhD-level mathematical research with the right orchestration.

In systems engineering, agents built a cycle-accurate, out-of-order RISC-V CPU simulator from scratch that boots the xv6 operating system to a shell, with 0.71% average cycle alignment error against BOOM hardware ground truth. In open-source software, Teamwork landed upstream performance work in Eigen, adding SIMD fast-paths for single-row or single-column GeMV, and in ParlayHash, where Swiss Parlay-style changes delivered 2x insert throughput with 64 threads and about 25% less memory per element.

Teamwork is not one fixed agent roster. Gemini picks a pattern from the prompt — Iterative Coding, Distributed Coding, Long Proof, Self-Verification, or Document Review — and the framework can change how many agents it spawns mid-run. The feature is available today as /teamwork-preview in Antigravity on all paid plans. Read more on the Google Keyword blog and the longer write-up on the Antigravity blog.