Can LLMs Correct Themselves? A Benchmark of Self-Correction in LLMs

Pan Zhou (Singapore Management University) · Junran Wu (National University of Singapore) · Lichao Sun (Lehigh University) · Zenghui Yuan (Hangzhou Dianzi University) · Ming Jin (Griffith University) · Qingsong Wen (Head of AI Research @ Squirrel AI) · Guiyao Tie (Huazhong University of Science and Technology) · Lixing Chen (Shanghai Jiaotong University) · Zeli Zhao (Huazhong University of Science and Technology) · Chaoran Hu (Huazhong University of Science and Technology) · Tianhe Gu (Huazhong University of Science and Technology) · Ruihang Zhang (Huazhong University of Science and Technology) · Sizhe Zhang (Huazhong University of Science and Technology) · Xiaoyue Tu (Huazhong University of Science and Technology)

chain-of-thought code generation commonsense reasoning complex reasoning taskscorrectbenchexternal methodsfine-tuned approachesintrinsic methodsmathematical reasoning optimization reasoning performance self-correction

Self-correction of large language models (LLMs) emerges as a critical component for enhancing their reasoning performance. Although various self-correction methods have been proposed, a comprehensive evaluation of these methods remains largely unexplored, and the question of whether LLMs can truly correct themselves is a matter of significant interest and concern. In this study, we introduce **CorrectBench**, a benchmark developed to evaluate the effectiveness of self-correction strategies, including intrinsic, external, and fine-tuned approaches, across three tasks: commonsense reasoning, mathematical reasoning, and code generation. Our findings reveal that: 1) Self-correction methods can improve accuracy, especially for complex reasoning tasks; 2) Mixing different self-correction strategies yields further improvements, though it reduces efficiency; 3) Reasoning LLMs (e.g., DeepSeek-V3) have limited optimization under additional self-correction methods and have high time costs. Interestingly, a comparatively simple chain-of-thought (CoT) baseline demonstrates competitive accuracy and efficiency. These results underscore the potential of self-correction to enhance LLM's reasoning performance while highlighting the ongoing challenge of improving their efficiency. Consequently, we advocate for further research focused on optimizing the balance between reasoning capabilities and operational efficiency.