ResearchBenchmarksReach-E on PlaygroundFollow918.9Average RewardFlow824.572849.061873.55898.039Mar 5, 2026Evaluation ResultsMethodMethodLinksAverage RewardFlowPolicy variant=diffusi...Policy variant=diffusion-only2026.03918.9FPO2026.03902.5Flow+ResPolicy variant=combine...Policy variant=combined optimal2026.03898.4DPPO2026.03828.2