Saltar al contenido principal

GPT 5.6 Luna Just Got 80% Cheaper: Do This Now

•4:47

Por , fundador de Genial Agency

AI Tool Review

You will know when GPT 5.6 Luna is the right choice and when Sol is, and you will have ChatGPT Work or Codex set up so Sol stays your main model and hands smaller jobs to Luna subagents at Max reasoning with fast mode.

Lo que necesitas

  • A ChatGPT subscription with access to GPT 5.6 Sol and Luna

  • ChatGPT Work or Codex (chatgpt.com/codex)

  • A project to apply the subagent setting to

Paso a paso

  1. Understand what changed

    GPT 5.6 Luna is now 80% cheaper. That makes it the second cheapest model at its level of intelligence, which I put alongside GLM 5.2, Opus 5 and Gemini 3.6. Your ChatGPT subscription now gets you more work for the same money.

  2. Look at the benchmarks

    In the benchmarks shown, Luna scores about the same as DeepSeek V4 Pro across the board. It also runs about three times faster (tokens per second) and costs far less. Treat that as a reason to try Luna, not as a reason to use it for everything.

  3. Decide when to use Sol and when to use Luna

    Think of the models as different people. Sol has the most intelligence and taste. Luna is strong at execution and matches Sol on small, easy tasks. Give Luna simple work, even when it is long. Keep harder, longer tasks on Sol.

  4. Keep design and UX on Sol

    Luna is still weaker than Sol at design and UX. If you are building your own app or vibe coding, use Sol.

  5. Use Sol as your main model with Luna subagents

    Here is the setup that gets you the most for your money: keep 5.6 Sol as your main model and let it send work to Luna subagents running with fast mode and the highest reasoning setting.

  6. Send the setup prompt in ChatGPT Work or Codex

    Open ChatGPT Work or Codex inside your project. Paste this prompt and send it, then wait for the reply.

    Setup prompt
    Hey ChatGPT, whenever you work with subagents, I want you to run them as Luna agents with the highest reasoning setting possible and fast mode. Can you set that up for me in this project?
  7. Choose where the setting applies

    You can apply this to one project, or repeat it for every project you have. Choose based on how you work.

  8. Check the setup and ask for Max reasoning

    Read the reply and check that everything you asked for is set up. Luna has a Max reasoning mode, but the settings menu for 5.6 Luna only goes up to Extra High. Ask for Max directly so your subagents run at full reasoning.

    Ask for Max mode
    For the Luna subagents, I need the max mode of reasoning plus fast mode.

Ten cuidado con

  • Luna being cheaper does not make it a good main model. It still falls behind Sol on longer, harder tasks.

  • Luna is weaker at design and UX, so keep app building and vibe coding on Sol.

  • The settings menu for 5.6 Luna stops at Extra High and does not show Max. You have to ask ChatGPT or Codex for Max yourself.

Qué ver después

Tutoriales relacionados

Formación: Capacitación práctica en IA para tu equipo

Empieza por una tarea o un proceso

En 30 minutos vemos cómo trabaja tu equipo y te decimos si conviene la formación, la automatización o ambas, y cómo se calcula el precio.