Development

A 27B AI scientist beats far larger models at reproducing unseen figures

development2026-08-16

Claim

Faraday, a 27-billion-parameter AI scientist trained to reproduce figures from papers it never saw, beat Opus 4.8 and GPT-5.5 in every category while wielding a bigger coding agent as its tool, and in 153 autonomous eight-day runs Claude Fable 5 closed 81.7% of the gap to a human speedrun record, though no run invented a new method.

Reported metric: 81.7% of the gap

Reporting date: (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Knowledge graphopen in browser →

Reported in
Evidence for
Supersedes
Referenced by

Cite or correct this record

Suggest a correction — opens a GitHub draft for you to review and submit.