DualDistill and Agentic-R1: How AI Combines Natural Language and Tool Use for Superior Math Problem Solving

Researchers from Carnegie Mellon University have introduced DualDistill, a novel framework that combines reasoning trajectories from two distinct teacher modelsone focused on natural language reasoning and the other on tool-augmented, code-based problem solvingto train a unified student model called Agentic-R1. This approach enables Agentic-R1 to dynamically select the most effective reasoning strategy for each problem, executing code for arithmetic and algorithmic tasks while relying on natural language reasoning for more abstract or conceptual challenges, thereby enhancing both efficiency and accuracy. By leveraging trajectory composition and self-distillation, DualDistill effectively merges the strengths of purely