ResearchBenchmarksLocal Question Answering on RealMMBench (TechReport)For AI assistantsMarkdownContent indexLocal Question Answering on RealMMBench (TechReport)Follow51.51AccuracyMegaRAG13.612423.451233.2943.1288Nov 26, 2025Evaluation resultsMethodMethodLinksAccuracyMegaRAGResponse LLM=GPT-4o-miniResponse LLM=GPT-4o-mini2025.1151.51LightRAGResponse LLM=GPT-4o-miniResponse LLM=GPT-4o-mini2025.1142.74NaiveRAGResponse LLM=GPT-4o-miniResponse LLM=GPT-4o-mini2025.1136.63GraphRAGSearch Mode=LocalSearch Mode=Local2025.1129.6GraphRAGSearch Mode=GlobalSearch Mode=Global2025.1115.07