On October 26, 2026, OpenAI announced the launch of three brand-new mathematics and science research results, and released a brand-new results release plan aimed at allocating AI-generated discoveries to academia. These breakthroughs cover multiple fields including theoretical computer science, mathematics, and quantum physics.
In terms of technical highlights, OpenAI's model made major progress in the field of computer science, claiming to have addressed relevant algorithm and optimization problems; in mathematics, by finding a new proof in all dimensions for the "local smoothing conjecture," it resolved the "Kakeya conjecture" in the study of waves and harmonics;
In the field of physics, the model proved a specific interacting quantum field theory containing four-dimensional gauge fields and fermions, covering a unified completion of quantum chromodynamics (QCD) and the Standard Model of particle physics. The related mathematical results had previously undergone formal verification in the proof assistant Lean, and some of the raw data has been uploaded to GitHub.
In response to the earlier controversy over obtaining results using undisclosed internal models, OpenAI adjusted its release strategy. The company pledged to no longer seek academic recognition for discoveries produced by internal models, instead positioning model-generated text as "raw research data" to be handed over to the mathematics and physics communities to be converted into scientific understanding.
To this end, OpenAI has contacted external experts to form teams as "guides," driving the conversion of AI output into human knowledge. However, regarding this new model, some scholars and industry insiders have pointed out that the selection of initial guides may bring potential challenges such as information asymmetry.
This progress marks a further deepening of the capabilities of large models in exploring the frontiers of fundamental science, and also reflects that AI companies are gradually exploring more standardized and cooperative paths of ecosystem governance when handing off machine-generated results to the scientific community.