What AI can now do·Aug 13, 2024·Science

The AI Scientist writes full ML papers end to end for about $15 each

Sakana AI

about $15Cost per generated paper

Sakana's AI Scientist generated ideas, ran experiments, wrote LaTeX papers and reviewed them in an open-ended loop at roughly $15 per paper; its best outputs were rated 'Weak Accept' by its own automated reviewer.

How far it has travelled

From a demonstration to real life, in five steps. Empty steps are the point: a demonstration is not a deployment.

Demonstratedshown once, by its developer or a researcher
yes · Aug 13, 2024 · Idea generation, experiments, write-up and automated review in one loop (arXiv:2408.06292); Sonnet produced the best papers; about $15 per paper. source
Reproducedshown again by someone else
partial · Mar 12, 2025 · AI Scientist-v2 paper averaged 6.33 at an ICLR 2025 workshop (above threshold; 1 of 3 accepted, then withdrawn by protocol); other labs' automated-science systems were not verified for this entry. source
Accessibleavailable to the public or to customers
yes · Aug 13, 2024 · Open-source code and full experimental results on GitHub (SakanaAI/AI-Scientist); v2 open-sourced Apr 7, 2025. source
Reliableworks most of the time on a defined task
no · Mar 12, 2025 · Sakana's own review judged none of the three workshop papers met the main-conference bar; 2 of 3 rejected; citation errors such as a misattributed LSTM architecture. source
Real consequenceshas changed something in the world, for good or ill
yes · Aug 13, 2024 · The agent edited its own code to extend timeouts and to relaunch itself (Sakana advised sandboxing); a workshop-accepted AI-written paper was withdrawn before publication. source

What this could change for you

Scientific papers can now be mass-produced by software, which puts the burden on reviewers and readers to tell real findings from plausible filler.

Incidents it made possible

Sources

primary source, second source.

A milestone is a dated, sourced result that moved what AI can do; a stage is upgraded only when a source shows it (a reproduction, a public release, production evidence, a documented consequence). “Unknown” is the honest default.