OpenAI says its artificial-intelligence models have produced a potential disproof of the Exponential Time Hypothesis, a long-standing conjecture in theoretical computer science. The announcement briefly confused parts of crypto social media because the hypothesis is commonly abbreviated as “ETH,” the same ticker used by Ethereum.
The distinction matters: OpenAI’s claim concerns the limits of algorithms, not the Ethereum blockchain, the price of Ether or Ethereum co-founder Vitalik Buterin. According to Decrypt, online reactions included jokes about algorithmic traders supposedly destroying ETH and references to Buterin, before users clarified that the acronym referred to a computer-science hypothesis.
The claim is now being examined by independent researchers, while OpenAI has stressed that the work requires substantial scrutiny and does not yet translate into a practical breakthrough for everyday computing.
What the Exponential Time Hypothesis says
The Exponential Time Hypothesis was proposed in 1999 by computer scientists Russell Impagliazzo and Ramamohan Paturi, together with Francis Zane. It concerns 3-SAT, a classic problem in which a computer must determine whether variables can be assigned true-or-false values so that a collection of rules, grouped three at a time, is satisfied.
ETH broadly proposes that 3-SAT cannot be solved substantially faster than searching through the possible combinations. Researchers have used the hypothesis in hundreds of papers to argue that certain problems in areas including graph theory, computational biology, databases and geometry are unlikely to have fast solutions.
OpenAI describes the hypothesis as a guide to what computers are unlikely to do efficiently. Its new paper argues that this boundary can be crossed. The 67-page paper, dated October 17 and revised October 25, presents a deterministic algorithm for counting every satisfying assignment in a 3-SAT problem in subexponential time. That means the amount of work increases more slowly than any exponential function of the number of variables.
Counting all solutions is a more demanding task than merely deciding whether at least one solution exists. Therefore, if the claimed algorithm is valid, it would also resolve the easier decision version in the same broad sense.
Researchers are checking the claim
OpenAI says its AI identified the purported disproof on October 16 and that the result was formalized in Lean the following day. Lean is software designed to check mathematical reasoning mechanically, allowing each logical step to be tested. The paper and formalization have been made available on GitHub.
Russell Impagliazzo, Ramamohan Paturi and MIT researcher Ryan Williams are leading an independent examination. In an open letter, the researchers cautioned that their participation should not be understood as certification. They said a result of this importance needs careful, independent scrutiny and must be reproducible, open to questions and usable by other researchers.
OpenAI said participants did not sign non-disclosure agreements and are working without payment. The company has also agreed to publish raw data and fund a conference focused on the result. Separately, OpenAI’s recent $5 million contribution to the Simons Foundation is intended to support workshops, conferences and special programs for examining and developing AI-generated research.
The company has warned against treating the paper as an immediate technological advance. Finding the algorithms is only one step, and converting theoretical progress into useful improvements will require significant additional research.
The broader computer-science implications
OpenAI’s paper does not resolve the famous P versus NP question. In simplified terms, P contains problems that computers can solve efficiently, while NP contains problems for which proposed answers can be checked efficiently. The unresolved question is whether every problem whose answer can be checked efficiently can also be solved efficiently.
The claimed result is narrower. The paper does not show that every NP problem has a subexponential-time solution, so the central P-versus-NP problem remains open.
Anthropic entered the discussion with a paper credited to Claude and titled “A 2^o(n) Time Algorithm for 3-SAT,” dated October 26. Its repository says an internal Claude research model discovered the algorithm, while also emphasizing that the result is purely asymptotic and does not demonstrate that the formulas can be solved practically.
The rapid appearance of overlapping AI-generated work has revived concerns about how research may change when models can investigate the same mathematical problem at high speed. Mathematician Terence Tao had previously warned that even the rumor of a research effort could prompt large amounts of AI-powered work before the original project had reached completion.
Moneyiar market picture
Moneyiar data recorded at 22:39 on October 25, 2026, showed Ethereum (ETH) at $2,342.0570017305, up 0.02% over 24 hours but down 1.77% over seven days. These figures refer to the cryptocurrency Ether and are separate from OpenAI’s use of “ETH” for the Exponential Time Hypothesis.
The same Moneyiar data showed Ethena (ENA) at $0.2092782867, up 8.19% in 24 hours and 12.35% over seven days. Polygon (POL), previously MATIC, was priced at $0.0898082186, up 0.39% over 24 hours and down 0.74% over seven days.
