Skip to content

Source paper

Towards Robust Mathematical Reasoning

Thang Luong, Dawsen Hwang, Hoang H. Nguyen and 17 more

Thang Luong, Dawsen Hwang, Hoang H. Nguyen, Golnaz Ghiasi, Yuri Chervonyi, Insuk Seo, Junsu Kim, Garrett Bingham, Jonathan Lee, Swaroop Mishra, Alex Zhai, Clara Huiyi Hu, Henryk Michalewski, Jimin Kim, Jeonghyun Ahn, Junhwi Bae, Xingyou Song, Trieu H. Trinh, Quoc V. Le, Junehyuk Jung

arXiv 2511.01846 · 20258 upvotes on Hugging Face

Research ideas from this paper (3)

Generated by AI. The paper’s authors did not write these.

AI summary

This paper introduces IMO-Bench, a challenging benchmark suite designed to evaluate mathematical reasoning at the International Mathematical Olympiad (IMO) level, comprising tasks for short-answer verification (IMO-AnswerBench), proof writing (IMO-ProofBench), and proof grading (IMO-GradingBench). The authors demonstrate that current frontier models, while improving, still struggle with robustification (reformulated problems) and subtle logical fallacies in proofs, proposing a new gold-standard model (Gemini Deep Think) that achieves significant gains through agentic search and verification.

How it works (AI summary)

The researchers curated a dataset of 400 short-answer and 60 proof-based Olympiad problems, applying 'robustification' techniques (e.g., variable renaming, adding distractors) to prevent memorization. They developed 'AnswerAutoGrader' and 'ProofAutoGrader' based on Gemini 2.5 Pro to evaluate model outputs, validating these graders against human expert ratings, and used an agentic framework with self-verification loops to achieve gold-medal level performance.