OpenAI publishes model-generated mathematics, Lean formalizations and process artifacts.
OpenAI publishes model-generated mathematics with Lean checks
2026-10-06Frontier watchWATCH
OpenAI published additional model-generated claims in harmonic analysis, complexity theory and quantum field theory under an external-shepherd model. It released the natural-language proof and Lean formalization for a claimed all-dimensions local-smoothing result, and a natural-language plus Lean refutation of the Exponential Time Hypothesis. OpenAI explicitly treats the manuscripts as raw research data; independent groups have not yet established broad acceptance.
Official · Referenceopenai.com · sharing ai progress in mathematicsOfficial · ReferenceOpenAI mathematics history
Availability: results public; producing model not public
What this means for you
For AI-for-science teams, the release shifts attention from model claims to public artifacts, dependency tracking and independent interpretation. Lean can check a formal statement against its axioms, but it does not by itself establish that the formal statement captures every intended scientific claim.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
OpenAI withdrew three manuscripts after a sign error, repaired fourteen and reported Lean coverage of 300/719 top-line results.
OpenAI publishes natural-language and Lean artifacts for local-smoothing/Kakeya and ETH claims while describing the results as raw research data awaiting academic interpretation.
Subsequent changes
OpenAI withdraws three model-generated math manuscripts and repairs fourteen
OpenAI publishes three new model-generated research claims with external shepherds
Technical details & scores
| Artifacts | papers, Lean formalizations, revision and citation protocol, 10 reasoning summaries |
|---|---|
| Reported Compute | roughly three hours of ChatGPT Pro thinking per result on average |
| Availability | results public; producing model not public |
| Verification Limit | formalization covers many, not necessarily all, proofs; expert review ongoing |
| Release Strategy | Math artifacts public; CS artifacts public with an external collaboration; physics artifacts held while collaborators prepare their own write-up |
| Verification Limit 2026 10 26 | OpenAI calls the manuscripts raw research data; broad independent acceptance is pending |
| Time Basis 2026 10 26 | OpenAI RSS published 2026-10-26T14:30:00Z; absent from the prior capture and recovered from the overlap |
Technical impact 4
Engineering value 3
Adoption signal 2
Maturity 2
Verification cost 4
Risk 3