ResearchBenchmarksRecipe Generation on Recipe Gen (test)Follow89.7Structural AccuracyRL-Struct43.73255.66667.679.534Nov 29, 2025Evaluation ResultsMethodMethodLinksStructural AccuracyJSON ValidityRL-StructAvg. Depth=3, learning...Avg. Depth=3, learning_paradigm=Reinforcement Learning2025.1189.792.1SFTAvg. Depth=3, learning...Avg. Depth=3, learning_paradigm=Supervised Fine-Tuning2025.1165.472.1Zero-shotAvg. Depth=3, learning...Avg. Depth=3, learning_paradigm=Zero-shot2025.1145.582.1