TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
OpenAI says its models produced research results in mathematics, computer science and physics, and it is testing different processes for academic groups to examine and explain them. The company has released the math writeup and Lean proof, while the computer science group has chosen to publish its natural-language proof and formalization; the physics collaborators are holding back the AI-produced materials for now. The claims and review process are still developing.
OpenAI says its models have produced research results in mathematics, computer science and physics, and it is testing three different ways for academic groups to examine and explain them. The company has released a mathematical proof and says its computer science collaborators chose to publish a proof and Lean formalization, while the physics group is preparing its own account before releasing the AI-produced artifacts.
The company describes the manuscripts as raw research data, rather than results that are already fully understood by their academic fields. It says outside experts were asked to act as “shepherds” and organize groups around each result. Their advice led to separate release approaches: publish the math output for researchers to assess, form a group to explain the computer science result while making its AI output available, and let the physics group guide the release of materials.
For mathematics, OpenAI says its models produced a proof of the local smoothing conjecture in every dimension, with implications for the Kakeya and Stein restriction conjectures. The company reports that it obtained the AI output on October 18, 2026, and formalized it in Lean the next day. It has made a natural-language writeup and Lean proof available publicly.
For computer science, OpenAI says its system produced a proof that the Exponential Time Hypothesis is false, and that the result was subsequently verified in Lean. The collaborating group decided to make a natural-language proof and Lean formalization public immediately; OpenAI says it is publishing those materials in its GitHub repository. For physics, the group has chosen to hold back the AI-produced artifacts while its members work on writing up the result themselves. The source material does not identify the physics result’s specific finding.
How the Release Choices Differ
The announcement concerns both the reported results and how AI-generated research enters scientific discussion. OpenAI’s three approaches assign different levels of control to outside researchers: the math output is open for independent examination, the computer science group is producing explanatory material alongside public artifacts, and the physics collaborators are leading the write-up before the AI materials are released.
That distinction matters because a technically detailed manuscript or formal proof may not, by itself, establish how a result fits into existing research or what follows from it. The computer science claim, if it withstands scrutiny, would affect a conjecture used to support conditional hardness results across fields including graph theory, computational biology and databases. OpenAI says further work would still be needed to determine whether the result leads to faster algorithms or practical improvements.
The process also raises questions about participation and credit. OpenAI says researchers take part voluntarily and without pay, do not sign nondisclosure agreements, and retain independence. It says the company is not seeking academic credit for discoveries made by its models, while collaborators should be recognized for the explanations and materials they produce. These are stated principles for the collaborations, not independent findings about their outcomes.
Products Worth Considering
Jackery HomePower 1000 Plus v2 Portable Power Station
Apple iPad Air (2025) 13‑inch Wi‑Fi 256 GB – Space Gray (M3 chip)
DJI Osmo Pocket 4 Gimbal Camera Pro Kit, 18-Piece Bundle
As an affiliate, we earn on qualifying purchases.
From Model Output to Peer Discussion
OpenAI says its approach has changed in response to feedback from academic communities. It characterizes the new release experiment as an attempt to turn model outputs into scientific understanding through expert discussion, explanation and contextualization. The company says it will provide collaborators with materials and artifacts related to the results.
The mathematical claim builds on earlier work described in the announcement. OpenAI notes that Sogge proposed the local smoothing conjecture in 1991; Guth, Wang and Zhang proved it in two spatial dimensions in 2019; and Wang and Zahl proved the three-dimensional Kakeya conjecture in 2025. The company says its claimed proof covers local smoothing in every dimension and is based on a novel auxiliary partial differential equation.
OpenAI also identifies tradeoffs in its own process: selecting initial shepherds can shape who participates first and create unequal access to information. It says it welcomes feedback on broadening participation. The announcement does not provide a full account of how the shepherds were selected or how later researchers can join each group.
Questions on Verification and Access
The announcement does not provide independent assessments of the three results or say that the relevant research communities have accepted them. OpenAI reports that the computer science result was verified in Lean, and that the mathematical output was formalized in Lean; formalization is distinct from field-wide review of the significance, assumptions and implications of a proof.
Several details remain unspecified. The source material does not name the physics finding, explain when its artifacts may be released, or set out a timeline for the computer science group’s explanatory work. It also does not give a complete process for joining the working groups or describe how disagreements between collaborators and OpenAI would be handled.
The claims about implications should be treated with care. OpenAI says the computer science result reopens questions about potential algorithmic improvements, but says further research is needed. Whether any such advances can be found, and whether they would translate into practical gains, is not established in the announcement.
Further Review and Group Write-Ups
The next step is for the academic groups to examine the results, develop explanations and publish their conclusions or other materials. For the computer science result, the natural-language proof and Lean formalization are being made public, giving researchers material to inspect and build on. For the physics result, collaborators say they are writing their own account before releasing the AI-produced artifacts.
OpenAI says it is seeking feedback on which release approach works best and how participation could be broadened. The company has not announced a schedule for further findings or a final assessment of the three strategies. The status of the claims will depend on continued examination by researchers and the release of additional details.
Key Questions
What did OpenAI announce?
OpenAI said its models produced three research results—in mathematics, computer science and physics—and described separate academic collaboration and release plans for each.
Has the computer science result been independently accepted?
The announcement says the Exponential Time Hypothesis result was verified in Lean and that collaborators chose to publish a proof and formalization. It does not say that the wider computer science community has accepted the claim.
What is being released for the mathematics result?
OpenAI says it has published a natural-language writeup and Lean proof for its claimed proof of the local smoothing conjecture in every dimension.
Why are the physics materials being held back?
OpenAI says the physics collaboration chose to defer release of the AI-produced artifacts while the group works on writing up the result. The announcement gives no release date or detailed explanation beyond that decision.
How will researchers take part?
OpenAI says participation in the collaborations is voluntary and unpaid, and that members can discuss the work and publish conclusions without nondisclosure agreements. It has not provided a complete process for additional researchers to join.
Source: openai.com
