Observed arrival · 2026-11-13
PhysDesignBench puts AI engineering designs through a rebuild
A public benchmark tests whether AI agents can turn physical specifications into valid engineering designs using real simulators.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Researchers evaluating AI engineering-design agents
- Worth noticing
- A failed design scores zero after independent checks for design rules, connectivity, electrical, thermal, and numerical limits.
Field notes
The page distinguishes Design tasks from Reimplementation tasks: the latter compare agents with published work, while the former ask them to produce a valid design from a specification. It reports that only one listed configuration found an eligible design on every Design task, and contrasts two similarly scoring configurations with different reported costs and run times. The homepage gives conflicting task totals: 26 in its metadata and 30 in its visible text.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
ƒJavaScriptBrowser-side code central
One card from the complete issue