OpenAI Shares New Mathematical Results from an Internal Frontier Model, Publishes Lean Proof Formalizations
OpenAI shared a broad range of new mathematical results produced by an internal frontier model, along with formal proof verifications, without yet releasing the model itself.
Details
- The model: results come from an unnamed internal frontier model that OpenAI says it is “working to responsibly release”; the model itself is not yet publicly available
- Formal verification: OpenAI is publishing formalizations of many of the proofs in Lean, a programming language that lets mathematical proofs be checked by computer, with plans to keep updating the repository over time
- Transparency materials: the release includes 10 summaries of the model’s reasoning process alongside statistics on the problems attempted
- Compute cost: OpenAI states the average result used roughly the equivalent compute of three hours of ChatGPT Pro “thinking”
- Oversight: the release and its citation/revision protocols were developed in consultation with the Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study
- Distribution: results and formalized proofs are published in a public GitHub repository
What happened next
The disclosure continues OpenAI’s pattern of pairing frontier-model capability claims with independently checkable evidence — in this case, machine-verifiable Lean proofs rather than self-reported benchmark scores — and signals that a more capable, not-yet-released model already exists internally ahead of whatever future launch OpenAI frames around it.