OpenAI publishes research-level proof attempts on the First Proof challenge

The internal model worked with limited human supervision; researchers sometimes requested retries or clarifications and selected among attempts.

What happened?

OpenAI published attempts on all ten First Proof problems and said at least five had a high chance of being correct, while explicitly retracting confidence in one earlier attempt after expert feedback.

Why it matters

The episode provided unusually transparent evidence of both research-level capability and the verification difficulty of AI-generated mathematics.

Significance rationale: First Proof moved frontier-model evaluation from contests and hidden benchmarks to expert-designed research proof attempts requiring prolonged community scrutiny.

Historical context

Primary evidence

Evidence & status at a glance

Significance
H2 · Field milestone
Verification
Under verification
Evidence level
E2 · Research record
AI role
AI primary, human verified
Formal assurance
No formal assurance

How should these labels be read? →

Contribution & verification

Mathematical novelty
Benchmark result
Interface
Informal, Symbolic
AI contribution
The model produced long research-level proof attempts across specialized domains.
Human contribution
Researchers prompted and selected attempts, experts supplied feedback, and correctness remained an explicit review question.
Verification
Under verification · E2 · Research record · No formal assurance
Last reviewed
2026-09-11

Technical details

Systems
openai-internal-reasoning-model
Organizations
openai, first-proof
Problems
first-proof
Methods
long-horizon-reasoning, expert-review, proof-generation

Sources & artifacts

Verification history

  • 2026-02-20 · Under verification · E2 · Research record

    The Chronicle records this event from the cited primary or official source; the status does not imply independent Chronicle verification.

Corrections

No corrections recorded.