Observed arrival · 2026-10-30
A one-mile channel, and the questions before crossing
Apocalypse Bench is a benchmark for testing whether AI models can solve physical-world problems by asking about hidden constraints.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- AI researchers and model evaluators
- Worth noticing
- The homepage says tasks have been solved in person and scores favor asking fewer questions.
Field notes
The described workflow gives a model a physical problem, a scene of unusual materials, and facts it must uncover by asking questions. The homepage links to separate Theory, Example, Findings, and Paper sections, but their contents are not present in the extracted text. It also states that every task has been solved in person, a useful grounding claim that cannot be independently assessed from this page extract.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
◎NicheUnusually specific use
One card from the complete issue