OpenAI Releases 722 AI-Generated Math Results From Frontier Model

OpenAI has released a major collection of mathematical research produced by one of its internal frontier AI models, highlighting how artificial intelligence is beginning to contribute to advanced mathematical research.

The company has published 722 mathematical manuscripts covering 372 groups of related results, with the work addressing difficult problems across multiple areas of mathematics. OpenAI says the release is intended to give mathematicians access to AI-generated discoveries while improving transparency around how these results were produced.

AI Tackles Difficult Mathematical Problems

The newly released work comes from an internal frontier model that OpenAI has been testing for advanced mathematical reasoning.

Rather than simply answering textbook-style equations, the model was used to investigate difficult research questions and generate potential mathematical results. OpenAI says the release includes solutions and discoveries spanning different areas of mathematics.

The company has also published additional information about the model's work, including summaries of its reasoning and statistics about the problems it attempted.

722 Manuscripts Cover 372 Result Families

One of the most notable aspects of the announcement is the scale of the release.

OpenAI says the 722 manuscripts represent 372 result families, meaning multiple manuscripts can be connected to the same broader mathematical result or line of research. The collection is intended to make the discoveries easier for researchers to examine, verify and build upon.

The company has published the material through a GitHub repository rather than keeping the research entirely behind a proprietary platform.

OpenAI Shares Details About AI Reasoning

OpenAI is also providing information about how much computational effort went into producing the mathematical results.

According to the company, the average result required computational effort equivalent to approximately three hours of ChatGPT Pro thinking. OpenAI has also released 10 summaries describing aspects of the model's reasoning process, along with statistics covering the number of mathematical problems attempted.

These details could help researchers better understand the relationship between AI reasoning time and the complexity of mathematical discoveries.

Many Proofs Can Be Checked With Lean

Another important part of the release is the use of Lean, a programming language and proof assistant that allows mathematical proofs to be formally checked by computers.

OpenAI says it is publishing formalizations of many of the proofs in Lean and plans to add additional formalizations as they become available. This could make it easier for mathematicians and computer scientists to verify whether AI-generated arguments are mathematically sound.

Computer-assisted verification is particularly valuable for AI-generated mathematics because even sophisticated AI systems can produce convincing-looking arguments containing subtle errors.

AI's Growing Role in Mathematical Research

The announcement represents another step in the rapidly developing relationship between artificial intelligence and mathematics.

AI models have already become increasingly capable at solving advanced mathematical problems. OpenAI previously reported that GPT-5.2 Thinking achieved a 40.3% score on FrontierMath Tier 1–3, an evaluation focused on expert-level mathematics.

The latest research release goes beyond benchmark scores by showing AI being used to generate mathematical results that researchers can examine and potentially extend.

OpenAI Wants Mathematicians to Build on the Results

OpenAI says it consulted the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study while developing its approach to releasing major mathematical results.

The company says the feedback helped shape its publication practices, including protocols for revisions and citations. OpenAI also plans to support workshops, conferences and other programs focused on understanding significant mathematical discoveries produced with AI.

This approach could help connect AI researchers with mathematicians who can independently evaluate and develop the results.

What This Means for AI Research

The release demonstrates that advanced AI systems are increasingly being evaluated not only on their ability to answer questions but also on whether they can contribute to scientific discovery.

If mathematicians verify and extend the newly published results, AI could become a powerful research assistant for areas involving complex proofs, theoretical mathematics and scientific computation.

However, the significance of the work will ultimately depend on independent verification and whether researchers can build meaningful new mathematics on top of the results.

Final Thoughts

OpenAI's release of 722 AI-generated mathematical manuscripts covering 372 result families marks a significant moment in the development of AI-powered research.

By publishing the work, reasoning summaries, computational details and formalized Lean proofs, OpenAI is giving mathematicians an opportunity to examine the discoveries directly.

The results still need to be carefully studied by experts, but the release suggests that frontier AI models are moving beyond solving established problems and toward becoming tools capable of contributing to new mathematical research.