OpenAI says its internal model produced 372 math breakthroughs, nearly all from a single prompt to one AI agent, though some may have taken multiple attempts

An AI that can independently solve major math problems, even if not yet fully understood, drastically lowers the barrier to complex discovery.

OpenAI announced that its internal AI model has generated 372 significant mathematical breakthroughs. This comes shortly after the company claimed its new large language model solved one of the six largest open problems in mathematics. The company stated that nearly all of these new results were produced by a single AI agent in response to a single prompt, a departure from its previous method involving a large swarm of agents. While many of the proofs have been verified using the Lean proof assistant, the AI model itself has not been publicly released. This lack of transparency has led to skepticism within the mathematics community, with some mathematicians calling for the release of the model, exact prompts, and compute times. OpenAI is currently working on releasing the model, balancing speed with responsible disclosure.

OpenAI's claim of a single-agent, single-prompt generation for 372 math breakthroughs signals a potential paradigm shift in AI-driven scientific discovery, moving from complex, costly "swarms" to highly efficient, solitary agents. This rapid generation, if reproducible, suggests AI could soon democratize advanced problem-solving, making sophisticated research tools accessible beyond large institutions. The speed at which these results are being produced also raises questions about the pace of scientific progress and the capacity of human researchers to integrate and validate AI-generated findings.

The skepticism surrounding OpenAI's unreleased model highlights a critical tension between rapid AI advancement and scientific rigor. The demand for transparency—release of models, prompts, and compute data—reflects a desire to ensure reproducibility and genuine understanding, not just high-volume output. This situation prompts a re-evaluation of how AI contributions to science are verified and integrated, potentially leading to new standards for AI research and disclosure in fields requiring verifiable proof, like mathematics.