--- id: "2026-11-13-google-research-antigravity-teamwork-gemini-4-argon-results" url: "https://postcutoff.com/e/2026-11-13-google-research-antigravity-teamwork-gemini-4-argon-results/" as_of: "2026-11-19T22:11:00+01:00" date: "2026-11-13" date_precision: day category: science importance: 4 confidence: medium status: [Partly confirmed] sources: 5 editor: human editor human_review: null version: "2026-11-17" --- As of: 2026-11-19 22:11 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/e/2026-11-13-google-research-antigravity-teamwork-gemini-4-argon-results/ # Google /teamwork: Weil in Lean, Courtade–Kumar, TCS results Full title: Google Research reports Antigravity /teamwork results with Gemini 4 Argon: Lean formalizations of the Weil conjectures (1.8M lines) and Hilbert's tenth problem over rings of integers, a 50M-line Lean proof of Courtade–Kumar, and a dozen TCS results On 13 Nov 2026 Google Research (post by Vahab Mirrokni) reported results from /teamwork, the multi-agent mode of Antigravity, run mostly on Gemini 4 Argon. In mathematics it lists a "fully autonomous" Lean formalization of the Weil conjectures through Deligne's 1974 proof (about 1.8M lines, 28,800 subagent invocations, via the LEAP framework), a 919K-line formalization of a recent proof of Hilbert's tenth problem over general rings of integers, a 50M-line Lean proof of the Courtade–Kumar conjecture, formalizations around 3D Navier–Stokes blow-up with smooth forcing, the 3D Bochner–Riesz conjecture (finding and fixing a gap in an informal proof) and Birkhoff's billiard conjecture, plus about a dozen algorithm and complexity results on arXiv. Gemini 4 models with /teamwork, /boost and Deep Review are now open to external users. All results are Google's claims. ## Key facts - Weil conjectures (Google): end-to-end machine-checked Lean formalization culminating in Deligne's 1974 proof, ~1.8M lines, 28,800 subagent invocations, via /teamwork-LEAP (arXiv 2606.03303); separate from Axiom's Weil formalization of 6 Nov - Hilbert's tenth problem over general rings of integers: 'over 919,000 lines of Lean' for the first formalization of a recent proof - Courtade–Kumar (2013): 'The team formalized the entire proof in Lean 4, producing 50 million lines of code'; Gemini 4 Argon and earlier models developed most of the analytical bounds and interval certificates and extended the proof to unbalanced functions (proof arXiv 2609.24931) - Other formalizations listed: 3D Navier–Stokes finite-time blow-up with smooth forcing (reproduced end to end), 3D Bochner–Riesz conjecture ('found and fixed a gap in a recently released informal proof'), Birkhoff conjecture (described as open since 1927), C¹ regularity of infinity-harmonic functions in dimension ≥ 3, Killip–Vişan KdV well-posedness in H⁻¹; checks of a UCLA group's proofs of the Jamison caterpillar and Ağırseven–Ollis conjectures and of a bound for Erdős #1063 - TCS results (arXiv): adaptive data analysis sample bounds, Lean-verified (2611.08699); Boolean functions needing more than n/2 quantum queries (2611.09809); deterministic combinatorial Boolean matrix multiplication resolving the STOC 2024 open question (2611.06717); Steiner-spanner lower bound (2608.08706); stacked sparse sketches with √log r nonzeros (2611.12867); single-pass cut sparsification (2611.03604); online inverse linear optimization (2609.26978); Tarjan's 1985 deque conjecture: splay-tree overhead below every iterated inverse Ackermann function, Lean-verified (GitHub sorrachai) - Deep Review at STOC 2027: authors of 578 submissions opted in; in a survey of 70, 90% rated the feedback mostly factual and over 30% said it caught substantive proof bugs or significant typos - Engineering: 800,000 lines of safe Rust completing a Zircon microkernel rewrite (boots on arm64); libgav1 rewritten in ~170k lines of Rust, 20–30% faster than the C++; 9×9 Go engine +2,600 Elo in under eight days; NMR structure identification 95% top-1 - Access: Gemini 4 with /teamwork, /boost and Deep Review open to external Antigravity users; six months of free top-tier access for up to 10,000 academic researchers ## What happened Google Research published a long progress report on /teamwork, the multi-agent mode of its Antigravity coding environment first described in August. The report mixes new mathematical and algorithmic results, very large Lean formalizations of known theorems, Rust rewrites and internal infrastructure savings, and announces external access for researchers. ## Why it matters It is the largest single batch of formalization claims from one lab so far: the Weil conjectures were formalized by two groups (Axiom on 6 Nov, Google here) within a week. The 50-million-line Courtade–Kumar proof also shows how far machine-written Lean can grow. The post itself does not give line-by-line evidence for every item; the arXiv papers and repositories are the primary sources. ## Your AI and this story - GPT-6.2 Astra (training cutoff May 2026): 166 days after its cutoff - Claude Opus 5.5 (training cutoff June 2026): 136 days after its cutoff - Gemini 4 Hydrogen (training cutoff March 2026): 227 days after its cutoff - Grok 4.7 (training cutoff May 2026): 166 days after its cutoff ## Sources 1. [Google Research blog: Advancing science & engineering with Antigravity's /teamwork (13 Nov 2026)](https://research.google/blog/advancing-science-engineering-with-antigravitys-teamwork-feature/) (research.google, official) 2. [LEAP framework (arXiv 2606.03303)](https://arxiv.org/abs/2606.03303) (arxiv.org, paper) 3. [arXiv 2611.06717: deterministic combinatorial Boolean matrix multiplication](https://arxiv.org/abs/2611.06717) (arxiv.org, paper) 4. [Tarjan deque conjecture bound in Lean (GitHub)](https://github.com/sorrachai/SplayTree-Deque-Ordinal-Inverse-Ackermann-Bound) (github.com, code) 5. [AI Weekly summary (14 Nov)](https://aiweekly.co/alerts/googles-antigravity-teamwork-multi-agent-system-formalizes-weil-hilberts-10th) (aiweekly.co, press) ## Changes - 2026-11-17 (filed): Created (science agent; Google Research post read in full) ## Related - 2026-11-17: [AI-assisted results, 17 Nov](https://postcutoff.com/e/2026-11-17-ai-assisted-results-november-17-listing/index.md) - 2026-11-06: [Axiom: the Weil conjectures proved in Lean](https://postcutoff.com/e/2026-11-06-axiom-weil-conjectures-lean-formalization/index.md) - 2026-10-14: [Google opens Deep Review, its agentic document-verification tool piloted on 16,000+ conference papers, to the public at Research@ 2026](https://postcutoff.com/e/2026-10-14-google-deep-review-public-launch-research-2026/index.md) - 2026-09-21: [Courtade–Kumar 'most informative Boolean function' conjecture (2013) proved three times in two days, all with AI: Ky & Tran (ChatGPT), Google + CUHK (Gemini, Lean-verified end-to-end), Mahdavifar & Beirami](https://postcutoff.com/e/2026-09-21-courtade-kumar-conjecture-proved/index.md) - 2026-08-27: [Google's Antigravity 'Teamwork' multi-agent framework with Gemini 3.7 Flash solves seven open CS/math problems, incl. part of Knuth's cycles problem](https://postcutoff.com/e/2026-08-27-antigravity-teamwork-open-problems/index.md)