ResearchBenchmarksReach-H on PlaygroundFollow873.4Average RewardFlow+Res736.848772.299807.75843.201Mar 5, 2026Evaluation ResultsMethodMethodLinksAverage RewardFlow+ResPolicy variant=combine...Policy variant=combined optimal2026.03873.4FPO2026.03870.1FlowPolicy variant=diffusi...Policy variant=diffusion-only2026.03857.2DPPO2026.03742.1