The AI Scientist v2, from Tokyo-based Sakana AI, is an open-source agentic system that automates scientific discovery end to end โ generating ideas, running experiments, analyzing results and writing up a workshop-level paper. It has around 7,400 stars on GitHub and was actively developed through 2025.
The system is built around agentic tree search: instead of following one fixed pipeline, an AI agent explores a tree of possible experiments, evaluates intermediate results and refines its approach, much like a researcher iterating on a problem. Starting from a broad research area, it proposes hypotheses, writes and executes experiment code, gathers and plots results, and drafts a full scientific manuscript describing what it found. Its predecessor made headlines as the first system to produce a peer-reviewed workshop paper written entirely by AI, and v2 generalizes the approach so it is not tied to specific experiment templates. The code is driven by modern foundation models for reasoning, coding and writing, and the repository includes everything needed to launch your own automated research runs. It is free and open source on GitHub.
The AI Scientist v2 is free and open source on GitHub. Running it costs whatever you spend on the underlying foundation-model API calls and compute for the experiments it launches โ there is no license fee for the system itself.
The AI Scientist v2 is for AI researchers curious about automated discovery, labs that want to prototype research-automation workflows, and anyone studying how far agentic systems can go in carrying out real scientific work.