AI Agent Protocol Benchmark: A Unified Framework for Evaluating Multi-Agent Communication
Published in arXiv preprint, 2025
This paper introduces a comprehensive benchmark for evaluating AI agent communication protocols across core tasks including Document QA, Collaborative Coding, and MAPF. The framework assesses protocols based on task performance, communication cost, and robustness to failures. Key contributions include protocol adaptation mechanisms, a Meta Protocol layer for unified interface integration, and fail-storm recovery experiments with Prometheus + OTLP monitoring for token-level cost and GPU usage tracking.
Recommended citation: Li, J., et al. (2025). "AI Agent Protocol Benchmark: A Unified Framework for Evaluating Multi-Agent Communication." arXiv preprint arXiv:2510.17149.
Download Paper