Morning Edition · Saturday, October 17, 2026Published at 2:12 AM EDT · New York

RealtimeBench shows a slower frontier model losing to a faster one once the game clock keeps running

The paper's teacher-student method raises the fast model's score from 13.51 to 24.19 out of 100 without calling the slow model during play.

Most agent benchmarks pause the environment while the model generates its response. A new paper, "The World Won't Wait", removes the pause. Its RealtimeBench contains 180 tasks across six game types. Under its live-clock protocol, the game…