Sakana AI launches AI Scientist v3: parallel coding agents automate research
hardmaru · x · 2026-10-17
Sakana AI released AI Scientist v3, "The Agentic Scientist." The third iteration shifts from "one agent writes a paper" to a scaffold of parallel coding agents orchestrating the full research loop.
Starting from a task.md, agents propose ideas, run experiments, evaluate outcomes against a task rubric, and iterate. The pipeline then runs baselines and ablations, a sealed final evaluation, post-hoc analyses, and paper writing.
Open-sourced under Apache-2.0 (SakanaAI/AI-Scientist-v3), with a blog post and arXiv paper accompanying the release.
Related event: Sakana AI Releases The Agentic Scientist, Paper Accepted at ALIFE 2026(10 posts)→
More from coding & agent
- AutoSep cuts agent reward hacking from 69.9% to 12.1% via weak-to-strong verifier hardening — Stanford · 2026-10-17
- Free open-source course teaches harness engineering with 14 lectures and 8 hands-on agent projects — techNmak · 2026-10-17
- Anthropic engineer: Claude Slides is now how I make slide decks — bcherny · 2026-10-17
- Dev built soft-body physics engine with Claude's help, claims it tops BeamNG — grandprixlegends · 2026-10-17
- tldraw ships Codex plugin for pair-programming on a live multiplayer canvas — max__drake · 2026-10-17
- Open Co-Scientist: open-source recreation of Google's AI Co-Scientist with Elo-ranked hypothesis evolution — Haunting-Stretch8069 · 2026-10-17