MTR-Bench: A Comprehensive Benchmark for Multi-Turn Reasoning Evaluation

TL;DR

MTR-Bench is a comprehensive, fully-automated benchmark designed to evaluate multi-turn reasoning capabilities of large language models across diverse, interactive tasks.

Contribution

It introduces a large-scale, multi-class, multi-turn reasoning benchmark with automated evaluation framework, filling a critical gap in interactive AI system assessment.

Findings

State-of-the-art models underperform on multi-turn reasoning tasks.

MTR-Bench covers 40 diverse reasoning tasks with 3600 instances.

Automated evaluation enables scalable, human-free assessment.

Abstract

Recent advances in Large Language Models (LLMs) have shown promising results in complex reasoning tasks. However, current evaluations predominantly focus on single-turn reasoning scenarios, leaving interactive tasks largely unexplored. We attribute it to the absence of comprehensive datasets and scalable automatic evaluation protocols. To fill these gaps, we present MTR-Bench for LLMs' Multi-Turn Reasoning evaluation. Comprising 4 classes, 40 tasks, and 3600 instances, MTR-Bench covers diverse reasoning capabilities, fine-grained difficulty granularity, and necessitates multi-turn interactions with the environments. Moreover, MTR-Bench features fully-automated framework spanning both dataset constructions and model evaluations, which enables scalable assessment without human interventions. Extensive experiments reveal that even the cutting-edge reasoning models fall short of multi-turn, interactive reasoning tasks. And the further analysis upon these results brings valuable insights for future research in interactive AI systems.