Codex Multi-Agent with Luna: Sol as Supervisor, Efficiently Batch Processing

According to Dynamic Beating monitoring, OpenAI has added a practical ability to Codex's multi-agent feature: Sol can now directly assign tasks to Luna. For example, the most capable Sol can be responsible for task decomposition and final inspection, and then multiple faster and cheaper Luna agents can work in parallel.

Previously, Codex already supported one agent calling multiple sub-agents and allowed different models to collaborate. However, Luna had a special restriction where it could only execute tasks on its own and could not be directly incorporated into a team by higher-level agents like Sol. OpenAI has now lifted this restriction.

OpenAI also explicitly recommends this division of labor. Sol is suitable for complex, open-ended tasks, while Luna is suitable for clear yet repetitive tasks such as searching, organizing, extracting, and categorizing. One is responsible for critical thinking and validation, while the others are responsible for bulk execution.