MetaGPT Releases New AI Development Capability Benchmark RealDevWorld

MetaGPT has launched a user agent, opening a new paradigm for end-to-end autonomous software testing. The agent has a dual role: acting both as a product manager conducting strict acceptance reviews and as a tireless AI test engineer, enabling full-process automation. The research team has released the RealDevWorld framework, which includes the RealDevBench dataset with 194 software development tasks and the evaluation agent AppEvalPilot.