Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2

2025/02/05 by Yuri Chervonyi, Trieu H. Trinh, Chervonyi, Yuri +20 · 18 voices · 36 citations
Computer Science · Mathematics · #Blockchain Technology in Education and Learning #Edcuational Technology Systems #Mathematics Education and Pedagogy #cs.AI #cs.LG

paper · pdf · doi:10.48550/arxiv.2502.03544

openalex publication_date 2025/02/05 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

We present AlphaGeometry2 (AG2), a significantly improved version of AlphaGeometry introduced in (Trinh et al., 2024), which has now surpassed an average gold medalist in solving Olympiad geometry problems. To achieve this, we first extend the original AlphaGeometry language to tackle problems involving movements of objects, and problems containing linear equations of angles, ratios, and distances. This, together with support for non-constructive problems, has markedly improved the coverage rate of the AlphaGeometry language on International Math Olympiads (IMO) 2000-2024 geometry problems from 66% to 88%. The search process of AG2 has also been greatly improved through the use of Gemini architecture for better language modeling, and a novel knowledge-sharing mechanism that enables effective communication between search trees. Together with further enhancements to the symbolic engine and synthetic data generation, we have significantly boosted the overall solving rate of AG to 84% on all geometry problems over the last 25 years, compared to 54% previously. AG2 was also part of the system that achieved the silver-medal standard at IMO 2024 https://deepmind.google/blog/ai-solves-imo-problems-at-silver-medal-level/. Finally, we report progress towards using AG2 as a part of a fully automated system that reliably solves geometry problems from natural language input. Code: https://github.com/google-deepmind/alphageometry2.

Citations

• Teaching Large Language Models to Self-Debug
• Generative Language Modeling for Automated Theorem Proving

Cited by

• Artificial Intelligence for Mathematical Reasoning: An Integrated Survey of Language Models, Neuro-symbolic Systems, and Verified Discovery
• Can LLMs Prove Robotic Path Planning Optimality? A Benchmark for Research-Level Algorithm Verification
• Formal Conjectures: An Open and Evolving Benchmark for Verified Discovery in Mathematics
• AlphaGo Moment for Model Architecture Discovery
• Winning Gold at IMO 2025 with a Model-Agnostic Verification-and-Refinement Pipeline
• Seed-Prover: Deep and Broad Reasoning for Automated Theorem Proving
• Hallucination Stations: On Some Basic Limitations of Transformer-Based Language Models
• Investigating the interaction of linguistic and mathematical reasoning in language models using multilingual number puzzles
• AutoHarness: improving LLM agents by automatically synthesizing a code harness
• AI4EOSC: a Federated Cloud Platform for Artificial Intelligence in Scientific Research
• Achieving Olympia-Level Geometry Large Language Model Agent via Complexity Boosting Reinforcement Learning
• Advancing Mathematical Research via Human-AI Interactive Theorem Proving
• GeoLoom: High-quality Geometric Diagram Generation from Textual Input
• Gold-Medal-Level Olympiad Geometry Solving with Efficient Heuristic Auxiliary Constructions
• Model-Based and Sample-Efficient AI-Assisted Math Discovery in Sphere Packing
• DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning
• Cognitive Foundations for Reasoning and Their Manifestation in LLMs
• Towards Robust Mathematical Reasoning
• GeoFM: Enhancing Geometric Reasoning of MLLMs via Synthetic Data Generation through Formal Language
• Symbolic Neural Generation with Applications to Lead Discovery in Drug Design
• Ax-Prover: A Deep Reasoning Agentic Framework for Theorem Proving in Mathematics and Quantum Physics
• An Alternative Trajectory for Generative AI
• ICL-Router: In-Context Learned Model Representations for LLM Routing
• The Idola Tribus of AI: Large Language Models tend to perceive order where none exists
• Efficient and Transferable Agentic Knowledge Graph RAG via Reinforcement Learning
• Semantic Voting: A Self-Evaluation-Free Approach for Efficient LLM Self-Improvement on Unverifiable Open-ended Tasks
• GenesisGeo: Technical Report
• How Large Language Models Need Symbolism
• Non-Parametric Structural Priors for Geometry Theorem Prediction
• Large language model-empowered next-generation computer-aided engineering
• The Need for Verification in AI-Driven Scientific Discovery
• Self-organized learning emerges from coherent coupling of critical neurons
• Discovering Expert-Level Nash Equilibrium Algorithms with Large Language Models
• LLM Unlearning Without an Expert Curated Dataset
• Goedel-Prover-V2: Scaling Formal Theorem Proving with Scaffolded Data Synthesis and Self-Correction
• AlphaGeometry [wikipedia]

Discussions

• OK! My Google colleague Thang Luong shared some exciting updates about AlphaGeometry2! AG2 now has surpassed the average gold-medalist in solving Olympiad geometry problems, w/ a solve rate of 84% co [bsky, 154 points, 1 comments]
• Gold-Medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 [hn, 64 points, 5 comments]
• These kinds of neurosymbolic systems – combining symbolic DSLs, search trees, and LLMs – are some of the more interesting things currently going on in AI imo. [bsky, 4 points, 0 comments]
• Google Deepmind 的 AlphaGeometry2 1. 第一次达到了 IMO 金牌选手的平均水平 2. 50道 2000年 到 2024年的 IMO 几何题目，OpenAI o1 一道都解决不了，而 AG2 可以解决 42 道，AG1 的水平是 27 道 3. “... many AlphaGeometry solutions to exhibit superhuman c [bsky, 3 points, 1 comments]
• AlphaGeometry from Deepmind achieved 84% solve rate on 2000-2024 International Mathematical Olympiad (IMO) geometry problems, surpassing gold medalist performance 🧮 made progress toward fully automa [bsky, 0 points, 0 comments]
• Gold-Medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 https://arxiv.org/abs/2502.03544 [bsky, 0 points, 0 comments]
• Gold-Medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 https://arxiv.org/abs/2502.03544 [comments] [39 points] [bsky, 0 points, 0 comments]
• 5️⃣ ¿Código abierto? Aún no se ha publicado, pero podría cambiar el panorama de la IA en matemáticas. 💡 ¿Cómo crees que afectará esto a la enseñanza y aplicación de las matemáticas? [bsky, 0 points, 1 comments]
• Paper: arxiv.org/abs/2502.03544 [bsky, 0 points, 0 comments]
• Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 arxiv.org/pdf/2502.03544 [bsky, 0 points, 0 comments]
• AlphaGeometry2 AI system achieving gold-medal level performance in solving complex Olympiad geometry problems, demonstrating advanced reasoning and mathematical problem-solving capabilities Read her [bsky, 0 points, 0 comments]
• Gold-Medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 (arxiv.org) Main Link | Discussion [bsky, 0 points, 0 comments]
• https://arxiv.org/abs/2502.03544 #cs.AI #cs.LG Event Attributes [bsky, 0 points, 0 comments]
• deep learning models seem far more promising and interesting to me than LLMs alphageometry2, for example, is actually insane: arxiv.org/abs/2502.03544 read that Ag1 was a mix of traditional algorith [bsky, 0 points, 0 comments]
• This by Yuri Chervonyi et al at DeepMind is fantastic but it's important to understand that this *isn't* "Gemini can do Olympiad-level geometry." It's "we improved our built-from-scratch specialized n [bsky, 0 points, 1 comments]
• Gold-Medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 https://arxiv.org/abs/2502.03544 (https://news.ycombinator.com/item?id=42969892) [bsky, 0 points, 0 comments]
• Gold-Medalist Performance in Solving Olympiad Geometry with AlphaGeometry2 https://arxiv.org/abs/2502.03544 (https://news.ycombinator.com/item?id=42969892) [bsky, 0 points, 0 comments]
• arxiv.org/abs/2502.03544 [bsky, 0 points, 0 comments]

Related

• Wu's Method can Boost Symbolic AI to Rival Silver Medalists and AlphaGeometry to Outperform Gold Medalists at IMO Geometry
• Gold-Medal-Level Olympiad Geometry Solving with Efficient Heuristic Auxiliary Constructions
• Proposing and solving olympiad geometry with guided tree search
• Achieving Olympia-Level Geometry Large Language Model Agent via Complexity Boosting Reinforcement Learning
• Newclid: A User-Friendly Replacement for AlphaGeometry