Cost-Aware Model Routing in Codex: Default to Luna, Escalate to Sol

Melvin Vivas · X post · 2026-08-18 · Open on X

Topics: LLM Fundamentals, AI Dev Tools & Productivity, LLMOps, Deployment & Monitoring · Level: intermediate

Summary

Compares two GPT-5.6 variants in Codex: Luna scores about 52 on intelligence at about $0.05 per task, and Sol scores about 61 at about $1.30 per task. Sol is 9 points smarter but costs 25x more. So the creator uses the cheap model by default and switches to the expensive one only when stuck.

Key points

• GPT-5.6 Luna: ~52 intelligence score, ~$0.05 per task
• GPT-5.6 Sol: ~61 intelligence score, ~$1.30 per task
• Sol gains ~9 points of intelligence for ~25x the cost
• Strategy: use the cheaper model for most coding tasks
• Switch to the stronger model only when the cheap one gets stuck
• Compare cost per task against capability before picking a default model

Resources mentioned

• GPT 5.6 Luna · tool · openai.com · paid
  The OpenAI model used inside Codex for the demo. The transcript gives the variant name as 'Soul', which is unclear.
  Also in: Set Codex subagent model and reasoning to save usage limits (Melvin Vivas on X · notes), Use GPT-5.6 Luna in Codex for Terminal Tasks (Melvin Vivas on X · notes), Match Reasoning Effort to Task Length in Codex (Astra/Sol) (Melvin Vivas on X · notes), Run Coworker desktop agents cheaply with GPT-5.6 Luna on OpenRouter (Melvin Vivas on X · notes) and 24 more
• GPT 5.6 Sol · tool · openai.com · paid
  A GPT-family model available through an API, whose API and credit pricing was cut by over 20% for three months.
  Also in: Creator's Top 3 Closed Models: Fable 5, GPT 5.6 Sol, Grok 4.6 (Melvin Vivas on X · notes), Coworker v0.3.1: Telegram Streaming Sync, Built Fast with Codex (Melvin Vivas on X · notes), Coworker: Open-Source Grok Bot Clone Built on Pi and CopilotKit (Melvin Vivas on X · notes), GPT 5.6 Sol (medium) in ChatGPT for planning tasks (Melvin Vivas on X · notes) and 12 more
• 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

Try this

• Default to GPT-5.6 Luna for most Codex tasks
• Switch to GPT-5.6 Sol only when Luna gets stuck