Soohak: A Mathematician-Curated Benchmark for Evaluating Research-level Math Capabilities of LLMs

TL;DR

Soohak is a new, large research-level math benchmark with 439 problems, designed to evaluate the reasoning capabilities of large language models, including their ability to recognize ill-posed problems.

Contribution

It introduces a comprehensive research-level math benchmark authored by mathematicians, including a refusal subset to assess models' ability to identify unsolvable problems.

Findings

Frontier models achieve around 30% accuracy on Challenge problems.

Leading open-weight models score below 15%.

Models struggle with recognizing ill-posed problems, with no model exceeding 50% on refusal subset.

Abstract

Following the recent achievement of gold-medal performance on the IMO by frontier LLMs, the community is searching for the next meaningful and challenging target for measuring LLM reasoning. Whereas olympiad-style problems measure step-by-step reasoning alone, research-level problems use such reasoning to advance the frontier of mathematical knowledge itself, emerging as a compelling alternative. Yet research-level math benchmarks remain scarce because such problems are difficult to source (e.g., Riemann Bench and FrontierMath-Tier 4 contain 25 and 50 problems, respectively). To support reliable evaluation of next-generation frontier models, we introduce Soohak, a 439-problem benchmark newly authored from scratch by 64 mathematicians. Soohak comprises two subsets. On the Challenge subset, frontier models including Gemini-3-Pro, GPT-5, and Claude-Opus-4.5 reach 30.4%, 26.4%, and 10.4% respectively, leaving substantial headroom, while leading open-weight models such as Qwen3-235B, GPT-OSS-120B, and Kimi-2.5 remain below 15%. Notably, beyond standard problem solving, Soohak introduces a refusal subset that probes a capability intrinsic to research mathematics: recognizing ill-posed problems and pausing rather than producing confident but unjustified answers. On this subset, no model exceeds 50%, identifying refusal as a new optimization target that current models do not directly address. To prevent contamination, the dataset will be publicly released in late 2026, with model evaluations available upon request in the interim.