Sakana AI open-sources Agentic Scientist v3, with ALIFE 2026 paper

Sakana AI has published its third version of the AI Scientist research-automation pipeline on GitHub under Apache 2.0, alongside an 88-page paper on arXiv submitted October 13, 2026.

"Turn a task description into experiments, ablations, and a paper, driven by coding agents," opens the project README. The system runs six phases in order (task drafting, investigation, ablations, final evaluation, post-hoc analysis, paper generation) and drives Claude, Codex, Cursor, OpenCode, Gemini and Pi through a Headless CLI.

The paper reports one real-world outcome.

The authors "leveraged a development version of our system to submit a single research paper, which got accepted for publication at ALIFE 2026." One acceptance, not a batch.

Robert Tjarko Lange, Luca Grillotti, Yutaro Yamada, Rujikorn Charakorn and David Ha pitch the system to run "across a spectrum of autonomy: either as a co-scientist tool or to perform end-to-end scientific investigation." The ablations single out the orchestration layer: "reliable scientific autonomy depends not only on base model capabilities, but also on the cross-agent orchestration layer."

Three of the researchers we track shared the link to the repository. Neither the README nor the arXiv abstract names the accepted ALIFE paper or publishes a head-to-head benchmark against the earlier AI Scientist versions.