The Agentic Scientist (AI Scientist v3)
Sakana AI's Apache 2.0 pipeline that orchestrates coding agents to turn a task.md file into experiments, ablations and a paper.
Overview
Sakana AI released the third generation of its AI Scientist, now called The Agentic Scientist, as the open-source SakanaAI/AI-Scientist-v3 repository under Apache 2.0, with an accompanying arXiv paper (2610.18933). Instead of its own agent loop, it drives existing coding agents through six sequential phases (task drafting, investigation, ablations, final evaluation, post-hoc analysis, paper writing) starting from a task.md file. Sakana reports that a development version co-authored an oral paper accepted at ALIFE 2026.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Six-phase pipeline from task.md to a written paper, with optional review and revision phases
- Works with claude, codex, cursor, opencode, gemini and pi through their headless CLIs
- agsci CLI with run, doctor, investigate, inspect and review commands
- Each phase runs in Docker by default; results written to one directory per run
- Web-based Run Inspector and optional AutoIllustrate for scientific figures (separate Apache 2.0 repo)
- Default budgets of 3 idea cycles, 5 ablation replications and 3 post-hoc analyses; Python 3.11+ required
- README warns agents run with full tool permissions and recommends using the Docker image
Use cases
- Automating experiment-and-write-up loops for ML research prototypes
- Benchmarking different coding agents on the same research task
- Reproducing GRPO-style training experiments (example preset needs 2 GPUs)
Pricing
Free and open source (Apache 2.0); you pay for the underlying coding-agent and model usage plus GPU time.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 15 Oct 2026
- Record updated
- 17 Oct 2026
- Record last checked
- 17 Oct 2026