ResearchBenchmarksPrompt Optimization on IFBenchFollow46.33ScoreLEVI35.815638.545341.27544.0047May 10, 2026Evaluation ResultsMethodMethodLinksScoreOptimization Budget (# Rollouts)LEVITask Model=Qwen 3 8BTask Model=Qwen 3 8B2026.0546.331,870GEPATask Model=Qwen 3 8BTask Model=Qwen 3 8B2026.0538.613,593BaselineTask Model=Qwen 3 8BTask Model=Qwen 3 8B2026.0536.9—MIPROv2Task Model=Qwen 3 8BTask Model=Qwen 3 8B2026.0536.22—