ResearchBenchmarksReinforcement Learning on RTFM two-monster, NL ShuffleReinforcement Learning on RTFM two-monster, NL ShuffleFollow83Average Win RateOracle MCTS525.2545.565.75Jun 21, 2026Evaluation resultsMethodMethodLinksAverage Win RateOracle MCTS2026.0683HRLLI2026.0629txt2πParameters=100MParameters=100M2026.0624Reader2026.0624ReaderArchitecture=TransformerArchitecture=Transformer2026.0623txt2π2026.068