ResearchBenchmarksStrategic game playing on Connect Four held-out (test)Strategic game playing on Connect Four held-out (test)Follow21.55Win RateMARSHAL10.037213.026116.01519.0039Oct 17, 2025Evaluation resultsMethodMethodLinksWin RateMARSHALConfiguration=Generali...Configuration=Generalist, Parameters=8B2025.1021.55Qwen3Parameters=8BParameters=8B2025.1010.48