ResearchBenchmarksFunction Calling on ToolUse (test)Function Calling on ToolUse (test)Follow91.2Structural AccuracyRL-Struct22.87240.61158.3576.089Nov 29, 2025Evaluation resultsMethodMethodLinksStructural AccuracyJSON ValidityRL-StructAvg. Depth=4, learning...Avg. Depth=4, learning_paradigm=Reinforcement Learning2025.1191.294.5SFTAvg. Depth=4, learning...Avg. Depth=4, learning_paradigm=Supervised Fine-Tuning2025.1170.178.5Zero-shotAvg. Depth=4, learning...Avg. Depth=4, learning_paradigm=Zero-shot2025.1125.545