[Codex Multi-Agent Adds Luna: Sol as Supervisor, Affordable Models Handle Tasks in Bulk] According to monitoring by Beating, OpenAI has added a practical capability to Codex's multi-agent functionality: Sol can now directly assign tasks to Luna. For example, the most capable Sol can handle task decomposition and final review, while multiple faster and more affordable Lunas work in parallel. Previously, Codex already supported one agent invoking multiple sub-agents and allowed different models to divide tasks. However, Luna had a specific limitation—it could only execute tasks independently and could not be directly integrated into a team by higher-level agents like Sol. This time, OpenAI has lifted that restriction. OpenAI now explicitly recommends this division of labor: Sol is suitable for complex, open-ended tasks, while Luna is ideal for clear and repetitive tasks such as searching, organizing, extracting, and classifying. One handles the thinking and validation, while a group handles bulk execution. [Original Link]