Nature npj Digital Medicine Publishes First Peer-Reviewed Benchmark of LLM Clinical Decision Agents

npj Digital Medicine·medium signal

A study in Nature npj Digital Medicine benchmarked multi-agent LLM systems for clinical decision tasks, evaluating OpenManus (Meta Llama-4 with medically customized agents) and Manus (proprietary planner-executor-verifier) across AgentClinic, MedAgentsBench, and Humanity's Last Exam benchmark families. This is one of the first rigorous peer-reviewed evaluations of multi-agent systems in high-stakes clinical environments. Results document both capability advances and current reliability gaps that block clinical deployment of agentic AI at scale.