In the field of relational databases,index selection is key to improving query performance.Accurately and efficiently estimating the benefits of candidate index configurations is crucial for query optimization.Traditional cost estimation methods based on "what-if" call tools suffer from low efficiency and insufficient accuracy,making it difficult to meet the requirements of large-scale and complex query workloads.Therefore,this paper proposes a Tree-based Transformer Index Benefit Estimation model (Tree-IBE).The model utilizes the self-attention mechanism of the Transformer architecture and efficient feature encoding techniques to directly learn index costs from query plans and index configurations,without relying on "what-if" call tools.Tree-IBE first performs feature encoding via a query plan node encoder and index encoder,comprehensively capturing key information from both.Then,it uses the multi-head attention mechanism of the Tree-Transformer layer to capture the dependencies between execution plan nodes in parallel across multiple representation subspaces.This allows it to accurately model the structure and performance characteristics of the execution plan before a lightweight estimation layer outputs the estimated index benefits.Finally,the model's performance is evaluated on three standard datasets:JOB,TPC-H,and TPC-DS.The results show that Tree-IBE outperforms existing methods in both accuracy and efficiency,particularly in complex query workloads.