ResearchBenchmarksDatabase Question Answering on EHRSQL (test)Follow35Execution AccuracyGATE26.05628.37830.733.022Jun 4, 2026Evaluation ResultsMethodMethodLinksExecution AccuracyGATEBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0635GATEBackbone=GPT-5.4-miniBackbone=GPT-5.4-mini2026.0634.4ReActBackbone=GPT-5.4-miniBackbone=GPT-5.4-mini2026.0632.2ReFoRCEBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0631.2ReActBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0630.2ReFoRCEBackbone=GPT-5.4-miniBackbone=GPT-5.4-mini2026.0627.6ReDelBackbone=GPT-5.4-miniBackbone=GPT-5.4-mini2026.0626.8ReDelBackbone=Qwen3-8BBackbone=Qwen3-8B2026.0626.4