Researchers proposed an Online MARL framework via one-step Flow model (OMAF) that combines expressive generative policies with efficient one-step action generation, reducing training overhead and improving sample efficiency in online multi-agent settings.