DeepAnalyze: Agentic Large Language Models for Autonomous Data Science

Autonomous data science has been a long-standing challenge, and is now becoming feasible with the emergence of powerful large language models (LLMs). Recent workflow-based data agents have shown promising results on specific data tasks but remain fundamentally limited in achieving full autonomy due to their reliance on predefined workflows. In this paper, we introduce DeepAnalyze, the first agentic LLM for autonomous data science, capable of automatically completing the end-to-end data science from raw data to research reports. To tackle high-complexity data science tasks, we propose a curriculum-based agentic training paradigm that emulates the learning trajectory of human data scientists, enabling LLMs to progressively acquire and integrate multiple capabilities in real-world environments. Accordingly, we contribute a data-grounded trajectory synthesis framework to constructs high-quality data science training data. Through training in real-world environment, DeepAnalyze learns to perform a broad spectrum of data tasks, ranging from data question answering to open-ended data research. Experiments on 13 benchmarks demonstrate that, with only 8B parameters, DeepAnalyze outperforms workflow-based agents built on advanced proprietary LLMs. DeepAnalyze is fully open-sourced, paving the way toward autonomous data science.

Data science often requires many steps: understanding raw data, asking useful questions, analyzing patterns, drawing conclusions, and writing up the results. Today, most automated tools can help with only parts of this process, usually by following fixed procedures designed in advance. This limits their ability to handle new, complex, or open-ended data problems. In this paper, we introduce DeepAnalyze, an open-source AI system designed to carry out the full data science process more independently. Starting from raw data, DeepAnalyze can explore the data, perform analyses, answer questions, and produce research-style reports. To make this possible, we train the system in a way that resembles how human data scientists learn: beginning with simpler tasks and gradually building toward more complex, real-world data investigations. We also develop a method for creating high-quality training examples grounded in actual data, helping the system learn from realistic data science workflows. Across a wide range of tests, DeepAnalyze performs better than previous automated data tools, even compared with systems built on larger proprietary AI models. By releasing DeepAnalyze openly, we hope to support further progress toward AI systems that can make data analysis more accessible, flexible, and useful for researchers and the public.