Saved in:
Bibliographic Details
Main Authors: Chen, Xiaoyang, Dai, Xinan, Du, Yu, Feng, Qian, Guo, Naixu, Gu, Tingshuo, Gao, Yuting, Gao, Yingyi, Han, Xudong, Jiang, Xiang, Jin, Yilin, Lin, Hongyi, Lin, Shisheng, Li, Xiangnan, Li, Yuante, Li, Yixing, Lai, Zhentao, Ma, Zilu, Peng, Yingrong, Qian, Jiacheng, Sun, Hao-Yu, Sun, Jianbo, Wang, Zirui, Wu, Siwei, Wang, Zian, Xu, Bin, Xu, Jianghao, Yu, Yiyang, Yang, Zichuan, Zha, Hongji, Zhang, Ruichong
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.08744
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916735487049728
author Chen, Xiaoyang
Dai, Xinan
Du, Yu
Feng, Qian
Guo, Naixu
Gu, Tingshuo
Gao, Yuting
Gao, Yingyi
Han, Xudong
Jiang, Xiang
Jin, Yilin
Lin, Hongyi
Lin, Shisheng
Li, Xiangnan
Li, Yuante
Li, Yixing
Lai, Zhentao
Ma, Zilu
Peng, Yingrong
Qian, Jiacheng
Sun, Hao-Yu
Sun, Jianbo
Wang, Zirui
Wu, Siwei
Wang, Zian
Xu, Bin
Xu, Jianghao
Yu, Yiyang
Yang, Zichuan
Zha, Hongji
Zhang, Ruichong
author_facet Chen, Xiaoyang
Dai, Xinan
Du, Yu
Feng, Qian
Guo, Naixu
Gu, Tingshuo
Gao, Yuting
Gao, Yingyi
Han, Xudong
Jiang, Xiang
Jin, Yilin
Lin, Hongyi
Lin, Shisheng
Li, Xiangnan
Li, Yuante
Li, Yixing
Lai, Zhentao
Ma, Zilu
Peng, Yingrong
Qian, Jiacheng
Sun, Hao-Yu
Sun, Jianbo
Wang, Zirui
Wu, Siwei
Wang, Zian
Xu, Bin
Xu, Jianghao
Yu, Yiyang
Yang, Zichuan
Zha, Hongji
Zhang, Ruichong
contents To advance the mathematical proficiency of large language models (LLMs), the DeepMath team has launched an open-source initiative aimed at developing an open mathematical LLM and systematically evaluating its mathematical creativity. This paper represents the initial contribution of this initiative. While recent developments in mathematical LLMs have predominantly emphasized reasoning skills, as evidenced by benchmarks on elementary to undergraduate-level mathematical tasks, the creative capabilities of these models have received comparatively little attention, and evaluation datasets remain scarce. To address this gap, we propose an evaluation criteria for mathematical creativity and introduce DeepMath-Creative, a novel, high-quality benchmark comprising constructive problems across algebra, geometry, analysis, and other domains. We conduct a systematic evaluation of mainstream LLMs' creative problem-solving abilities using this dataset. Experimental results show that even under lenient scoring criteria -- emphasizing core solution components and disregarding minor inaccuracies, such as small logical gaps, incomplete justifications, or redundant explanations -- the best-performing model, O3 Mini, achieves merely 70% accuracy, primarily on basic undergraduate-level constructive tasks. Performance declines sharply on more complex problems, with models failing to provide substantive strategies for open problems. These findings suggest that, although current LLMs display a degree of constructive proficiency on familiar and lower-difficulty problems, such performance is likely attributable to the recombination of memorized patterns rather than authentic creative insight or novel synthesis.
format Preprint
id arxiv_https___arxiv_org_abs_2505_08744
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DeepMath-Creative: A Benchmark for Evaluating Mathematical Creativity of Large Language Models
Chen, Xiaoyang
Dai, Xinan
Du, Yu
Feng, Qian
Guo, Naixu
Gu, Tingshuo
Gao, Yuting
Gao, Yingyi
Han, Xudong
Jiang, Xiang
Jin, Yilin
Lin, Hongyi
Lin, Shisheng
Li, Xiangnan
Li, Yuante
Li, Yixing
Lai, Zhentao
Ma, Zilu
Peng, Yingrong
Qian, Jiacheng
Sun, Hao-Yu
Sun, Jianbo
Wang, Zirui
Wu, Siwei
Wang, Zian
Xu, Bin
Xu, Jianghao
Yu, Yiyang
Yang, Zichuan
Zha, Hongji
Zhang, Ruichong
Artificial Intelligence
To advance the mathematical proficiency of large language models (LLMs), the DeepMath team has launched an open-source initiative aimed at developing an open mathematical LLM and systematically evaluating its mathematical creativity. This paper represents the initial contribution of this initiative. While recent developments in mathematical LLMs have predominantly emphasized reasoning skills, as evidenced by benchmarks on elementary to undergraduate-level mathematical tasks, the creative capabilities of these models have received comparatively little attention, and evaluation datasets remain scarce. To address this gap, we propose an evaluation criteria for mathematical creativity and introduce DeepMath-Creative, a novel, high-quality benchmark comprising constructive problems across algebra, geometry, analysis, and other domains. We conduct a systematic evaluation of mainstream LLMs' creative problem-solving abilities using this dataset. Experimental results show that even under lenient scoring criteria -- emphasizing core solution components and disregarding minor inaccuracies, such as small logical gaps, incomplete justifications, or redundant explanations -- the best-performing model, O3 Mini, achieves merely 70% accuracy, primarily on basic undergraduate-level constructive tasks. Performance declines sharply on more complex problems, with models failing to provide substantive strategies for open problems. These findings suggest that, although current LLMs display a degree of constructive proficiency on familiar and lower-difficulty problems, such performance is likely attributable to the recombination of memorized patterns rather than authentic creative insight or novel synthesis.
title DeepMath-Creative: A Benchmark for Evaluating Mathematical Creativity of Large Language Models
topic Artificial Intelligence
url https://arxiv.org/abs/2505.08744