How do I run GPT-6 Luna in Codex or OpenCode?

Copy a working config

Codex is the natural home for Luna, and OpenCode works with the same key. Pinning Luna keeps every request on it; leave the model as standardcompute if you want harder steps to go to a stronger model you selected.

# Install Codex and the Standard Compute launcher once
npm install -g @openai/codex
curl -fsSL https://standardcompute.com/cli/install.sh | sh

# Start Codex pinned to openai/gpt-6-luna
standardcompute launch codex -- --model openai/gpt-6-luna

{
  "$schema": "https://opencode.ai/config.json",
  "model": "standardcompute/openai/gpt-6-luna",
  "provider": {
    "standardcompute": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Standard Compute",
      "options": {
        "baseURL": "https://api.stdcmpt.com/v1",
        "apiKey": "{env:STANDARDCOMPUTE_API_KEY}"
      },
      "models": {
        "openai/gpt-6-luna": { "name": "GPT-6 Luna" },
        "standardcompute": { "name": "Smart routing" }
      }
    }
  }
}

Replace YOUR_API_KEY with a key from the dashboard; OpenCode reads it from STANDARDCOMPUTE_API_KEY. Full setup: Codex CLI, OpenCode.

What it costs

OpenAI standard API rates. Regional data-residency endpoints add 10%. Source: OpenAI API pricing, checked October 13, 2026.

At that rate the $20 budget in the $19 Starter plan goes a long way on Luna alone. A plan pays off when Luna shares the month with heavier models.

What works differently here

1. Codex's built-in web search is off on this connection, and the launcher refuses --config, --profile and --oss so it can keep the connection settings it writes.
2. The Responses API here is stateless: previous_response_id returns 400. Codex sends the full conversation and works as shipped; a custom Responses client must do the same.
3. Luna is the efficient tier. GPT-6 Sol is not on Standard Compute yet, so for harder steps select GPT-6 Astra or GPT-5.6 Sol next to Luna and let routing decide.

Other ways to run GPT-6 Luna

FAQ