ResearchBenchmarksVisual Question Answering on DocVQA T1Follow78.6EMPoP69.44871.82474.276.576Feb 27, 2026Evaluation ResultsMethodMethodLinksEMHallucination RateAverage Budget Usage (B-bar)PoPReasoning protocol=PoPReasoning protocol=PoP2026.0278.67.111.2ProgVLMReasoning protocol=Pro...Reasoning protocol=ProgVLM2026.027511.913.9MM-ReActReasoning protocol=MM-...Reasoning protocol=MM-ReAct2026.0274.312.714.8M-CoTReasoning protocol=M-CoTReasoning protocol=M-CoT2026.0272.115.40.25DirectReasoning protocol=DirectReasoning protocol=Direct2026.0269.817.90.25