Observed arrival · 2026-11-11
ParkBench’s AI theme-park stress test
ParkBench compares eight AI models running the same RollerCoaster Tycoon park for 16 simulated months in OpenRCT2.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Readers of AI evaluations and simulation-based agent benchmarks
- Worth noticing
- The report tracks operational behavior too: one model issued 3,962 park commands, nearly three times another model’s 1,418.
Field notes
The experiment gives each owner a park image every two weeks of game time and tracks guests, rating and cash as the run unfolds. Its findings connect outcomes to concrete choices: one owner checked individual map squares 1,581 times, while another left food stalls inaccessible and ended with 333 of 372 guests lost.
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
One card from the complete issue