Arcs Directions
Back to Timeline

Event record

OpenAI Reports Ten Mathematics Breakthroughs Solved or Advanced by Internal Astra Model

OpenAI announced that an unreleased internal version of Astra solved 10 open math problems, including three Erdős conjectures, releasing Lean formal certificates.

Event Summary

On August 1, 2026, OpenAI announced that an unreleased internal version of its Astra model solved or made substantial progress on 10 long-standing problems in mathematics and theoretical computer science, including three Erdős conjectures. OpenAI released Lean formal certificates and model reasoning narrations for each proof.

Context & Narrative

Following its May 2026 disproof of the Erdős unit distance conjecture, OpenAI announced ten additional advances in pure mathematics and theoretical computer science produced by an unreleased internal version of its Astra model. The results span high-dimensional geometry, group theory, quantum complexity, and extremal combinatorics, resolving Erdős problems 146, 180, and 183. For each proof, OpenAI published Lean formal certificates and model reasoning logs on GitHub. Human researchers assisted in preparing manuscripts and checking formalizations.

Key Findings

  • Fact Grade B

    On August 1, 2026, OpenAI announced that an internal unreleased version of Astra solved or substantially advanced 10 open problems in mathematics and theoretical computer science, including 3 Erdős conjectures.

    Sources [1][2]
  • Fact Grade B

    OpenAI published Lean formal certificates and model reasoning narrations for all 10 proofs in an open code repository.

    Sources [1][3]
  • Impact Grade C

    Demonstrates machine generation of formally verified mathematical proofs for human-curated open conjectures, raising benchmark standards for AI in pure mathematics.

    Sources [2]
  • Limitation Grade B

    The Astra model remains unreleased to the public, and human researchers assisted in preparing manuscripts and formalization; long-term mathematical significance remains subject to peer evaluation.

    Sources [1][2]

Impact Assessment

  • Capability Leap +2 · Long-term

    Demonstrates machine generation of formally verified mathematical proofs for human-curated open conjectures.

    Affected Groups: mathematicians, computer scientists, AI researchers

Consensus & Sources

Significance L1
Category Capability Breakthrough
Consensus Emerging Consensus
Impact Index 5/10