Cutting Codex Subagent Costs: Luna Orchestrator, Astra Reviewer

Melvin Vivas · X post · 2026-09-07 · Open on X

Topics: AI Agents, Tool Use & MCP, AI Dev Tools & Productivity, LLMOps, Deployment & Monitoring · Level: intermediate

Summary

Melvin Vivas suggests a cheaper Codex multi-agent setup for Plus users. Usage was high mainly because the expensive Astra model ran as the root orchestrator at high reasoning effort. His fix is to make GPT-5.6 Luna at max reasoning the orchestrator and use Astra only as a reviewer.

Key points

• Heavy usage in the Astra-Luna subagents setup likely came from Astra running as the root orchestrator with reasoning effort set to high by default.
• Cheaper option: set the root orchestrator in config.toml to model = "gpt-5.6-luna" with model_reasoning_effort = "max".
• Keep the stronger Astra model only for the reviewer role, so it runs less often.
• Pick each role's model and reasoning effort on purpose, because these settings drive cost and usage limits.
• The creator said a guide on how to measure usage was coming.

Resources mentioned

• OpenAI Codex · tool · openai.com · paid
  OpenAI's coding agent. In the diagram it writes code, fixes review findings and drives the build loop. The creator also used it to make this video.
  Also in: An agent bot that installs and drives Codex on its own (Melvin Vivas on X · notes), Asking a Coder bot to install Codex (Melvin Vivas on X · notes), Sign in with ChatGPT: Setting Usage Limits for Each App (Melvin Vivas on X · notes), Codex Cloud Environments Must Be Saved & Published Before Use (Melvin Vivas on X · notes) and 240 more
• donvito/codex-astra-luna-orchestrator · repo · github.com · free
  The creator's Codex skill for an orchestrator-plus-subagents setup (Astra orchestrator, Luna subagents), configured through config.toml.
  Also in: Codex Orchestrator Pattern: Astra/Sol Orchestrator with Luna Subagents (Melvin Vivas on X · notes), Open-source Codex skill: Astra/Sol orchestrator with Luna subagents (Melvin Vivas on X · notes), Cost-Efficient Codex Subagents: Astra/Sol Orchestrator + Luna Workers (Melvin Vivas on X · notes), Codex Orchestrator v0.2.1: GPT-6 Sol orchestrator with Luna subagents (Melvin Vivas on X · notes) and 27 more

Try this

• If usage is high on a Plus plan, set the root orchestrator to gpt-5.6-luna with model_reasoning_effort = "max" and keep Astra as the reviewer.
• Check what reasoning effort your orchestrator model uses by default.