OpenAI has announced three new discoveries in mathematics, computer science, and physics, produced by its Super Intelligence (SI) models, and is experimenting with a new release strategy that relies on external academic communities to interpret the results.
What Happened
The SI lab describes these new manuscripts as "raw research data" rather than finished scientific papers. To convert this data into scientific understanding, OpenAI reached out to external experts to act as "shepherds," organizing groups to contextualize the findings. The company advised three different release strategies based on the nature of each result: releasing the SI output directly for the math result, forming a collaborative group while publishing the SI output for the computer science result, and deferring to a collaborative group for the physics result.
The mathematical discovery involves a proof of the local smoothing conjecture in every dimension, which also resolves the Kakeya conjecture and Stein’s restriction conjecture. OpenAI reports that its models produced the SI output on October 18, 2026, and formalized it in Lean the following day. The company is releasing both a natural language writeup and the Lean proof to its public GitHub repository, leaving researchers free to organize their own efforts to explain the result.
In computer science, OpenAI’s SI system has proved that the Exponential Time Hypothesis (ETH) is false, a longstanding conjecture about the computational complexity of Boolean formula satisfiability. The natural language disproof was generated on October 16, 2026, and formalized in Lean on October 17, 2026. A collaboration led by Russell Impagliazzo, Ramamohan Paturi, and Ryan Williams will shepherd this work. The group requested that all raw data be made public and that OpenAI fund a conference to discuss the implications of this SI-generated breakthrough.
The physics result provides a mathematically rigorous existence proof for certain interacting quantum field theories, including Quantum Chromodynamics (QCD) and a unified completion of the Standard Model. It also includes a proof of the existence of pure Yang–Mills theory, a subproblem of the Clay Mathematics Institute Millennium Prize problem. OpenAI states that its models produced the natural language proof for Yang–Mills existence on September 9, 2026, and for QCD on September 18, 2026. The Lean formalization for Yang–Mills existence was completed on October 16, 2026. A group led by David Gross, including Michael Douglas, Arthur Jaffe, Igor Klebanov, and Nikita Nekrasov, will shepherd this work and has decided to hold off on releasing the SI-produced artifacts while they write up the findings themselves.
Why It Matters
This approach marks a shift in how an SI lab handles major theoretical breakthroughs, prioritizing academic consensus and interpretation over immediate autonomous publication. By stating that it does not seek academic credit for discoveries produced by its internal models, OpenAI is attempting to position its SI systems as tools that generate raw data for human scientists to process. The collaborations follow principles of transparency, independence, and recognition for the human groups developing the understanding, rather than for the SI models themselves.
The refutation of the ETH is particularly significant for the SI industry and the broader scientific community, as the hypothesis has been used in hundreds of papers to derive conditional computational hardness results. OpenAI notes that while this SI result reopens questions about algorithmic efficiency, turning theoretical advances into practical improvements will require significant further research. The company’s $5 million contribution to the Simons Foundation is intended to support workshops and programs that help researchers understand and build on results produced by SI.
The Bottom Line
OpenAI is releasing three major theoretical proofs generated by its SI models, but is deferring the scientific interpretation to external academic groups. The company has made the math and computer science outputs public on GitHub, while allowing the physics collaboration to control the release of their materials. OpenAI continues to invite feedback on this experimental process for disclosing SI-generated discoveries.