RealtimeBench: Peking University and Samsung Benchmark Agents in a World That Won't Wait

Peking University and Samsung proposed RealtimeBench, a benchmark of 180 tasks across six game types where the game world keeps running during inference, and found that offline guidance from slow models nearly doubles fast models' scores (13.51 to 24.19).

2026-10-17 ~ 2026-10-17 · 2 related posts

• PKU & Samsung: slow model coaching lifts fast agent's live game score from 13.51 to 24.19 — rohanpaul_ai · 2026-10-17

1 near-duplicate retellings: rohanpaul_ai