ResearchBenchmarksExecutable World-Model Learning on Default World (held-out extra-level transitions)Executable World-Model Learning on Default World (held-out extra-level transitions)Follow100Overall AccuracyAlice86.06489.68293.396.918May 16, 2026Evaluation resultsMethodMethodLinksOverall AccuracyBalanced AccuracyAliceEvaluation Protocol=Of...Evaluation Protocol=Offline, Calls=22, LLM Backbone=GPT-5.42026.05100100CWMEvaluation Protocol=Of...Evaluation Protocol=Offline, Calls=100, LLM Backbone=GPT-5.42026.0596.292.8GIF-MCTSEvaluation Protocol=Of...Evaluation Protocol=Offline, Calls=100, LLM Backbone=GPT-5.42026.0586.679.9