ResearchBenchmarksCard Prediction on HigherLower 100 episodes (test)Follow0.53RewardSLM0.49880.50690.5150.5231Jul 2, 2026Evaluation ResultsMethodMethodLinksRewardAccuracyIRORSLMModel=4BModel=4B2026.070.5374100—ASK+Model=2BModel=2B2026.070.5274696ASK+Model=4BModel=4B2026.070.5274605SLMModel=2BModel=2B2026.070.5173100—PPOModel=—Model=—2026.070.5720—ASKModel=2BModel=2B2026.070.573690ASKModel=4BModel=4B2026.070.572600