The Multi-step Task Success Rate is a critical evaluation metric for Agents tasks in the fields of large language models and reasoning. It represents the proportion of instances in which an agent successfully executes all sub-tasks within a multi-step task and achieves the final goal, reflecting the agent's long-term planning and robust execution capabilities. This metric is typically reported as a percentage or a score between 0 and 1. It is widely used in relevant benchmarks and papers, effectively reflecting model performance in this dimension. Higher values are better. It helps researchers and engineers quantify the differences in actual model performance in Agents scenarios, facilitating horizontal comparisons across different methods or model versions.