ResearchBenchmarksReward Modeling on ManiRewardBench Bimanual YAMFollow94.7Mean VOCTOPReward-3.0622.3247.773.08Feb 22, 2026Evaluation ResultsMethodMethodLinksMean VOCTOPRewardBackbone=Qwen3-VL-8BBackbone=Qwen3-VL-8B2026.0294.7GVLBackbone=Gemini-2.5-Pr...Backbone=Gemini-2.5-Pro, Chat template enforced=true2026.0256.6TOPRewardBackbone=Molmo2-8BBackbone=Molmo2-8B2026.0256.5TOPRewardBackbone=Gemini-2.5-Pr...Backbone=Gemini-2.5-Pro, Chat template enforced=true2026.0254.6GVLBackbone=Qwen3-VL-8BBackbone=Qwen3-VL-8B2026.0216.4GVLBackbone=Molmo2-8BBackbone=Molmo2-8B2026.020.7