ResearchBenchmarksPrompt Optimization on HotpotQAPrompt Optimization on HotpotQAFollow63ScoreLEVI41.503247.084152.66558.2459May 10, 2026Evaluation resultsMethodMethodLinksScoreOptimization Budget (# Rollouts)LEVITask Model=Qwen 3 8BTask Model=Qwen 3 8B2026.05632,750GEPATask Model=Qwen 3 8BTask Model=Qwen 3 8B2026.0562.336,871MIPROv2Task Model=Qwen 3 8BTask Model=Qwen 3 8B2026.0555.33—BaselineTask Model=Qwen 3 8BTask Model=Qwen 3 8B2026.0542.33—