The Confidence Dichotomy: Analyzing and Mitigating Miscalibration in Tool-Use Agents

Benchmark Model Rank Results
mathematical-reasoning-on-aime-2025Qwen2.5-3B with CAR (MSCR)–Accuracy: 21.1
mathematical-reasoning-on-aime24Qwen2.5-3B with CAR (MSCR)–Acc: 20.8
mathematical-reasoning-on-math-500Qwen2.5-3B with CAR (MSCR)–Accuracy: 76.9