AnswersPublished12 sources
What AlphaProof Nexus’s math results show—and what they don’t
A May preprint reports 9 solutions among 353 Erdős problems and 44 proofs among 492 OEIS conjectures—about 2.5% and 8.9%, respectively.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- A May preprint reports 9 solutions among 353 Erdős problems and 44 proofs among 492 OEIS conjectures—about 2.5% and 8.9%, respectively.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: What does Google DeepMind’s October 8, 2026 Science paper, following its May arXiv preprint, report about AlphaProof Nexus’s solutions to ni. Article summary: The May preprint reports a meaningful but selective advance: AlphaProof Nexus resolved nine of 353 Erdős problems and proved 44 of 492 OEIS conjectures, with reported computing costs of a few hundred dollars per solved E. Topic tags: general, academic, general web, user generated, government. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, wate
A May 2026 arXiv preprint reports that Google DeepMind’s AlphaProof Nexus resolved nine of 353 open Erdős problems and proved 44 of 492 conjectures from the Online Encyclopedia of Integer Sequences (OEIS). The reported computing cost was a few hundred dollars per solved Erdős problem. The figures point to a potentially useful proof-search system, but they are not, by themselves, a measure of how reliably it can do mathematics or whether every result is new. 17
What the reported results amount to
The preprint’s totals work out to about 2.5% of the Erdős problems and 8.9% of the OEIS conjectures tested. Those are calculated shares of the evaluated sets, not success rates for every mathematical question the system might encounter. The authors also report that experts checked whether each successful Erdős-problem Lean statement faithfully represented the original conjecture. 17
Those distinctions matter: the paper reports selected successes from much larger collections, not a general guarantee that the system can solve open problems on demand. And the stated cost is per solved Erdős problem; on its own, it doesn’t establish the total cost of the unsuccessful attempts or the cost-effectiveness of the full search process. 17
Why formal verification is important—but not the whole story
AlphaProof Nexus is presented in the preprint as a framework for AI-driven formal proof search. In this kind of workflow, a proof is expressed in a formal system such as Lean so that a proof checker can verify it against a formal statement. A checked proof can provide strong assurance that the formal argument follows from that statement. 17
But verification and mathematical context are separate questions. A checker’s acceptance does not by itself establish that a formalized statement matches the intended open problem, that the result has not already been proved, or that the result is significant. The preprint’s report that experts checked the correspondence between the Lean statements and the Erdős conjectures is therefore an important part of interpreting the headline count. 17
What the available evidence does—and doesn’t—confirm
An October news report says that two of the reported Erdős results concern questions posed by Paul Erdős and András Sárközy in 1970. That supports the claim that the achievements include long-standing problems, but it is secondary coverage rather than the paper itself. 19
The sources available here include the May preprint and secondary reports, but not the full October Science paper or the detailed independent assessments needed to evaluate every claim in the question. They therefore do not establish the named model version, the reported algebraic-geometry and min-max results, or the specific disputes about prior solutions, changed wording, alternative agents, placeholder proofs, and human-directed searches. Those details should not be treated as settled without checking the paper, its formal proof artifacts, and the relevant reviewers’ accounts.
How to assess the achievement
A careful assessment is problem by problem: check the original conjecture, confirm that its formal version matches, inspect the complete proof, review prior literature, and document any human involvement. The reported totals make AlphaProof Nexus worth examining as a proof-search tool. They do not, on their own, settle whether it independently conducts mathematical research or how novel and important each result is. 17
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "What AlphaProof Nexus’s math results show—and what they don’t"?
A May preprint reports 9 solutions among 353 Erdős problems and 44 proofs among 492 OEIS conjectures—about 2.5% and 8.9%, respectively.
Sources
arxiv.orgAdvancing Mathematics Research with AI-Driven Formal ...
elmundo.esGoogle DeepMind logra que una IA resuelva problemas matemáticos que llevaban más de 50 años sin solución
science.orgAt futuristic meeting, AIs took the lead in producing and reviewing all the studies