ResearchBenchmarksEmbodied Agent Action Prediction on ALFWorldFollow65AccuracyDPO+54.657.36062.7Jun 25, 2026Evaluation ResultsMethodMethodLinksAccuracyDPO+Backbone=Qwen3-8BBackbone=Qwen3-8B2026.0665SFT+Backbone=Qwen3-8BBackbone=Qwen3-8B2026.0663FRODOBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0661DPOBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0659SFTBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0655