---
title: "AdvancedMathBench: A Benchmark Suite for Advanced Mathematical Proof Generation and Verification"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.11849v1"
permalink: "https://feed.astralor.com/items/itm_74f28fd8f092e3f6e33cd039"
published: "2026-07-13T17:38:22.000Z"
score: 78
channel: "ai"
tags: ["AdvancedMathBench","GPT-5.5","LLM","Mathematical Reasoning","Proof Verification"]
---

# AdvancedMathBench: A Benchmark Suite for Advanced Mathematical Proof Generation and Verification

## AI commentary

> What strikes me most is the fine-grained evaluation of advanced mathematical reasoning—no longer just final answers, but the proof process and error detection. For researchers aiming to push LLM reasoning boundaries, ProverBench and VerifierBench offer valuable testing grounds. Especially relevant for AI reasoning researchers and evaluation teams.

## At a glance

AdvancedMathBench is a benchmark suite comprising ProverBench (296 problems at undergraduate and doctoral qualifying-exam levels) and VerifierBench (888 model-generated proof trajectories with expert annotations) to evaluate LLMs on advanced mathematical proof generation and verification. Experiments show that GPT-5.5 achieves only 75.8/66.1 on UGD/QE splits for generation and a Balanced F1 of 65.1 on verification, with low true negative rates, indicating critical error detection as a major bottleneck.

## Key points

- Introduces AdvancedMathBench, including ProverBench and VerifierBench, covering advanced mathematics from undergraduate to doctoral qualifying-exam levels.
- Develops an automatic proof verification pipeline trained on large-scale expert annotations, showing strong agreement with human experts.
- GPT-5.5 achieves only 75.8/66.1 on UGD/QE splits for proof generation, and a Balanced F1 of 65.1 on proof verification.
- Models exhibit low true negative rates, making critical error detection a major bottleneck.

## Full text

Large language models (LLMs) have achieved remarkable performance on high-school and olympiad-style mathematics, yet their capabilities on advanced mathematics remain poorly understood. Existing benchmarks, however, fall short in both scope and evaluation granularity: they provide limited disciplinary coverage and often rely on final-answer correctness or coarse judgments, leaving the validity of the reasoning process inadequately assessed. To bridge this gap, we introduce AdvancedMathBench, a benchmark suite designed to evaluate advanced mathematical reasoning capabilities. Its core proof-generation benchmark, ProverBench, contains 296 problems spanning undergraduate and doctoral qualifying-exam levels. To provide reliable evaluation of the proofs, we develop a dedicated automatic verification pipeline trained on large-scale expert annotations to produce both correctness verdicts and fine-grained assessments of proof errors, which exhibits strong agreement with human experts on held-out proof trajectories. We further introduce VerifierBench, consisting of 888 model-generated proof trajectories paired with expert ground truth, to evaluate whether models can correctly judge proof validity and provide sound verification rationales. Experiments show that AdvancedMathBench remains challenging for frontier models. On proof generation, the best-performing model, GPT-5.5-xhigh, achieves only 75.8 and 66.1 on the UGD and QE splits, respectively, indicating substantial room for improvement on advanced mathematical proof construction. On proof verification, the best model attains a Balanced F1 of only 65.1, and models generally exhibit low true negative rates, suggesting that critical error detection remains a major bottleneck.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_74f28fd8f092e3f6e33cd039