Autonomous agents built on large language models need common protocols to interoperate across systems. Several such standards have emerged — A2A, MCP, ACP, ANP, and NLIP — but the Natural Language Interaction Protocol (NLIP) has not appeared in prior benchmarks. A new arXiv preprint, "The Cost of a Hop: Benchmarking NLIP and A2A," sets out to fill that gap.
The paper evaluates NLIP against A2A, focusing on the "cost of a hop" — the overhead incurred when an agent makes a single interaction or request across the protocol. By quantifying this cost, the authors aim to give developers a clearer picture of the trade-offs between different interoperability standards.
Because this is a single preprint, the findings are not yet peer-reviewed, and the abstract alone does not reveal the benchmark's specific results or methodology. Still, the work highlights a practical concern: as agent ecosystems grow, the choice of protocol may have measurable performance implications.