From Thoughts to Trees: Multi-Path Reasoning and Tool Integration in LLMs

Nov 1, 2025
  • Delivered 12-18% accuracy improvement over CoT and ReAct on GSM8K, ARC, and LogiQA.
  • Added parallel branch scoring and pruning.
  • Reduced incorrect tool calls by 25%.
Llms Reasoning Agents
Abhijit Chunduru
Abhijit Chunduru
Master’s Student

ML Systems and Trustworthy AI at UMass Amherst