ResearchBenchmarksReward Prediction on 10-task benchmark OverallReward Prediction on 10-task benchmark OverallFollow0.02Demo L (MSE)SARM20.016520.040010.06350.08699Jun 9, 2026Evaluation resultsMethodMethodLinksDemo L (MSE)SARM2standard_full=truestandard_full=true2026.060.02SARM2Multi-Goal=excludedMulti-Goal=excluded2026.060.026SARM2Spectral Embedding=exc...Spectral Embedding=excluded2026.060.034RW2026.060.036RobometerFine-tuning=trueFine-tuning=true2026.060.043RobometerFine-tuning=falseFine-tuning=false2026.060.093TR2026.060.107