English

When Agents Disagree With Themselves: Measuring Behavioral Consistency in LLM-Based Agents

Artificial Intelligence 2026-02-13 v1

Abstract

Run the same LLM agent on the same task twice: do you get the same behavior? We find the answer is often no. In a study of 3,000 agent runs across three models (Llama 3.1 70B, GPT-4o, and Claude Sonnet 4.5) on HotpotQA, we observe that ReAct-style agents produce 2.0--4.2 distinct action sequences per 10 runs on average, even with identical inputs. More importantly, this variance predicts failure: tasks with consistent behavior (\leq2 unique paths) achieve 80--92% accuracy, while highly inconsistent tasks (\geq6 unique paths) achieve only 25--60%, a 32--55 percentage point gap depending on model. We trace variance to early decisions: 69% of divergence occurs at step 2, the first search query. Our results suggest that monitoring behavioral consistency during execution could enable early error detection and improve agent reliability.

Keywords

Cite

@article{arxiv.2602.11619,
  title  = {When Agents Disagree With Themselves: Measuring Behavioral Consistency in LLM-Based Agents},
  author = {Aman Mehta},
  journal= {arXiv preprint arXiv:2602.11619},
  year   = {2026}
}

Comments

5 pages, 2 figures

R2 v1 2026-07-01T10:33:06.274Z