NetGoodIndexSubmit a correction

Benefit Ledger · verified · Mathematics

AlphaGeometry solves olympiad plane-geometry problems without human proof data

DeepMind’s neuro-symbolic AlphaGeometry solved 25 of 30 IMO-AG-30 geometry problems in a Nature paper, approaching average human gold-medalist performance on that set.

17 Jan 2024Tier 3 MajorMethodology 0.1

Current score

+2.77

10 base · Major (tier 3 of 5, 10 pts)
× 0.9000 attribution · Primary causal contribution
× 0.7000 evidence · Peer review or independent validation
× 0.5500 realization · Experimentally validated
× 0.8000 durability
Event-level product before credit split: 2.77

Major automated-reasoning result on a defined benchmark (tier 3). High attribution to the system. Peer-reviewed evaluation, not live contest medals. Realization is demonstrated on a historical set. Durability is high for methods and proofs but the specific model is replaceable.

What happened

On 17 January 2024, Nature published AlphaGeometry, a language model trained on synthetic geometry theorems that guides a symbolic deduction engine. On 30 recent olympiad-level plane-geometry problems it solved 25, versus 10 for a prior automated baseline. Human-readable proofs were produced; former olympiad medalist Evan Chen reviewed examples. The system covers only Euclidean plane geometry, not a full IMO paper.

Model attribution

+2.77

AlphaGeometry

Neuro-symbolic prover that proposed auxiliary constructions for a symbolic engine.

The synthetic-data language model plus deduction engine is the published system.

Attribution 0.9000 · Credit share 100% · Google DeepMind

Claims

  • AlphaGeometry solved 25 of 30 problems in the IMO-AG-30 geometry set.

    outcome · supported

  • AlphaGeometry by itself earned an IMO gold medal in a live contest.

    outcome · disputed

Sources

primary sources

Secondary domains: Computer Science

Revision history

  • 13 Sep 2026 · 0.00 → 2.77

    Initial adjudicated seed score under methodology 0.1.