GPT-5.2 derives new gluon amplitude formula in physics first

GPT-5.2 has proposed a new formula for a gluon amplitude, according to a preprint — a result that was later formally proved and verified by OpenAI and academic collaborators.

The preprint describes a workflow in which the model generated a candidate expression for a gluon amplitude, a mathematical object describing the scattering probability of gluons, the force-carrying particles of the strong nuclear force. Human researchers then took that candidate through formal proof and independent verification. OpenAI worked alongside academic collaborators to confirm the result.

This is the sequence that matters. The model did not simply reproduce a known result from its training data. It proposed something new in theoretical physics, and that proposal survived the standard scrutiny of the field — formal proof and verification by domain experts. The preprint documents the derivation and the subsequent confirmation.

Gluon amplitudes sit at the technical core of quantum chromodynamics, the theory of the strong interaction. Progress in computing amplitudes has historically come from physicists working through specialized algebraic techniques. A general-purpose language model contributing a new formula here marks a shift in what these systems can do in domains where answers cannot be checked by intuition alone — they must be proved.

The verification step is the load-bearing part of the claim. In theoretical physics, a proposed formula has no standing until it is formally derived and checked. The fact that OpenAI and its academic collaborators completed that process, and that the work now appears in preprint form, means the result has entered the public record for the research community to examine.

For AI research, the significance is methodological. The reported result demonstrates a model moving beyond pattern-matching on known answers toward generating candidates for genuinely new science, with humans retaining control of proof and validation. For physics, it suggests a new kind of collaborator: one that can propose formal structures faster than they can be checked, shifting the bottleneck toward verification.

The claim will face the usual scrutiny that greets any preprint, and the formal proof and verification described by OpenAI and its academic collaborators will be the evidence the community weighs. If the result holds up under that examination, it will strengthen the case that frontier models can contribute original mathematics to physics rather than only summarize it.