VAR-MATH: Probing True Mathematical Reasoning in LLMs via Symbolic Multi-Instance Benchmarks

Published in arXiv, 2025

This work introduces VAR-MATH, a symbolic multi-instance benchmark that probes whether LLM mathematical gains reflect genuine reasoning rather than benchmark-specific overfitting or memorization.

Recommended citation: Yao, J., Zheng, B., Cheng, R., and Tan, K. C. (2025). "VAR-MATH: Probing True Mathematical Reasoning in LLMs via Symbolic Multi-Instance Benchmarks." arXiv:2507.12885.
Download Paper