The Data Interpreter is an LLM-based agent designed to enhance data science problem-solving by integrating dynamic planning, tool utilization, and automated confidence-based verification. It addresses challenges in data science, including real-time data adaptability, complex task dependencies, and logical error identification. The agent employs hierarchical graph structures for dynamic planning, enabling real-time adjustments and efficient task management. It also integrates tools and generates custom functions to improve coding proficiency and adaptability. Additionally, it uses automated verification to ensure logical consistency and accuracy in code execution. The Data Interpreter outperforms existing open-source frameworks in various data science tasks, including machine learning, mathematical problems, and open-ended tasks. It achieves a 10.3% improvement in machine learning tasks, a 26% increase in the MATH dataset, and a 112% improvement in open-ended tasks. The agent's performance is validated through extensive experiments, demonstrating its effectiveness in handling complex data science challenges. The Data Interpreter's key contributions include a dynamic planning framework with hierarchical structures, automated tool integration, and experience-driven reasoning. These components enhance adaptability, coding proficiency, and reasoning accuracy. The agent's ability to learn from past experiences and adjust to new data makes it highly effective in data science scenarios. Overall, the Data Interpreter represents a significant advancement in LLM-based agents for data science, offering improved performance and reliability in complex problem-solving tasks.The Data Interpreter is an LLM-based agent designed to enhance data science problem-solving by integrating dynamic planning, tool utilization, and automated confidence-based verification. It addresses challenges in data science, including real-time data adaptability, complex task dependencies, and logical error identification. The agent employs hierarchical graph structures for dynamic planning, enabling real-time adjustments and efficient task management. It also integrates tools and generates custom functions to improve coding proficiency and adaptability. Additionally, it uses automated verification to ensure logical consistency and accuracy in code execution. The Data Interpreter outperforms existing open-source frameworks in various data science tasks, including machine learning, mathematical problems, and open-ended tasks. It achieves a 10.3% improvement in machine learning tasks, a 26% increase in the MATH dataset, and a 112% improvement in open-ended tasks. The agent's performance is validated through extensive experiments, demonstrating its effectiveness in handling complex data science challenges. The Data Interpreter's key contributions include a dynamic planning framework with hierarchical structures, automated tool integration, and experience-driven reasoning. These components enhance adaptability, coding proficiency, and reasoning accuracy. The agent's ability to learn from past experiences and adjust to new data makes it highly effective in data science scenarios. Overall, the Data Interpreter represents a significant advancement in LLM-based agents for data science, offering improved performance and reliability in complex problem-solving tasks.