Observed arrival · 2026-11-18
GEPHYRA tests whether AI agents notice—and act
A research project building benchmarks for AI agents in tasks they’re asked to do and situations where they must spot a problem themselves.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- AI-agent researchers and model-evaluation teams
- Worth noticing
- Partner choices are scored against what happens when the models are actually trained.
Field notes
GEPHYRA-Select focuses on a partner-selection problem: organizations share summaries, and an agent chooses whom to work with. The project says those choices will be judged against training outcomes, rather than only against a suitability description. The page identifies the benchmark as under construction and says code, data, and results will be released on the site, Hugging Face, and PyPI; it gives no release date.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
◉HumanPersonal, local, civic, or handmade
One card from the complete issue