DeepAnalyze is an agentic large language model built for autonomous data science, developed by the Data Lab at Renmin University of China. It can take a data task from raw tables to a finished analytical report without human intervention, and the project has earned more than 4,700 stars on GitHub in 2026.
Described by its authors as the first agentic LLM for autonomous data science, DeepAnalyze handles the entire workflow an analyst would follow: understanding the request, exploring and cleaning data, writing and running code, building visualisations and compiling findings into a professional report. The team has released the DeepAnalyze-8B model on Hugging Face together with the DataScience-Instruct-500K training dataset, so the system can be studied, fine-tuned and self-hosted. It works across databases and notebook-style Python environments and is aimed at the broad range of data-centric tasks found in research and business analysis, from quick exploration to deep-research style investigations.
The project’s website and paper describe evaluations across data analysis scenarios, and the release of both the specialised model and its instruction dataset makes DeepAnalyze one of the more complete open stacks for studying autonomous agents. Rather than acting as a chat layer over existing tools, it is trained to behave as the analyst itself, deciding which operations to perform and in what order.
DeepAnalyze is free and open source under the MIT license. The code, model weights and dataset are publicly available; you only provide the compute to run the model or the API budget for models you connect.
DeepAnalyze is for data teams, researchers and analysts who want an AI data analyst that does the work rather than just answering questions. It suits organisations exploring automated reporting and researchers studying agentic models specialised for data-centric reasoning.