How to Connect OpenAI Models to OpenClaw

API key, Codex login or Sign in with ChatGPT?

All three use the same provider name, openai, and the same model references such as openai/gpt-6-astra. What changes is who pays and which extra OpenAI tools work.

Connect with your ChatGPT account (Codex login)

openclaw models auth login --provider openai

openclaw models auth login --provider openai --method device-code

openclaw config set agents.defaults.model.primary openai/gpt-6-astra

openclaw models list --provider openai
# then, in a chat with your assistant:
/codex status
/status

Connect with an OpenAI Platform API key

export OPENAI_API_KEY="sk-..."
openclaw onboard --openai-api-key "$OPENAI_API_KEY"

openclaw onboard --auth-choice openai-api-key

openclaw models auth login --provider openai --method api-key

openclaw models list --provider openai
openclaw models set openai/gpt-6-astra
openclaw models status --probe --probe-provider openai

Sign in with ChatGPT (SIWC)

SIWC authorizes OpenClaw as its own app on your ChatGPT account and spends your Codex allowance. Your account and workspace must have SIWC and token sharing enabled by OpenAI.

openclaw models auth login --provider openai --method siwc

1. Approve token sharing during sign-in. If you approve identity only, OpenClaw saves the account but can't run models with it.
2. The browser returns to http://localhost:8080/auth/callback. If your browser is on another computer, forward the port first:
   On your browser's computerCopy

   ssh -N -L 8080:127.0.0.1:8080 user@gateway-host

3. Check shared usage at ChatGPT Settings → Usage. OpenClaw doesn't show SIWC quota.

Choose a GPT model

Run openclaw models list --provider openai to see what your account can use. API projects and ChatGPT plans can offer different models.

*US dollars per million tokens at standard API rates, as listed in the OpenClaw docs. Check OpenAI's pricing page for current rates. GPT-6 models accept text and images and have a 1,050,000-token context window. OpenClaw uses 272,000 tokens of active input by default. For images, use openai/gpt-image-2.

OpenAI config builder

Check your real profile names with openclaw models auth list --provider openai and adjust the auth.order list to match. With a backup key, OpenClaw tries the plan first, then the key.

Set the reasoning effort

GPT-6 models default to medium effort. Change it from any chat:

/think high      # harder problems
/think default   # back to medium
/think ultra     # lets OpenClaw split big jobs across sub-agents
/fast on         # Fast mode (API key, higher price)

• /think off works on GPT-6 Sol and Luna. Astra and GPT-6.1 Sol can't turn reasoning off.
• /think ultra isn't a raw API setting. It's OpenClaw's orchestration mode, running at max effort.
• A saved minimal setting becomes low.

Images, voice and memory search

Not every OpenAI tool works with every sign-in. Here's what each one covers, according to the OpenClaw docs:

Use OpenAI embeddings for OpenClaw's memory search:

{
  memory: {
    search: {
      provider: "openai",
      model: "text-embedding-3-small",
    },
  },
}

Cost and usage tracking

Fix common OpenAI connection problems

openclaw models status
openclaw models auth list --provider openai
openclaw doctor --fix
openclaw config validate

More fixes: troubleshooting guide.

OpenAI and OpenClaw questions