OpenClaw on your ChatGPT plan: setup, model choice and limits (and why Claude plans work differently)

Yes, OpenClaw runs on a ChatGPT Plus or Pro plan. Run openclaw onboard --auth-choice openai (or openclaw models auth login --provider openai) and sign in with your ChatGPT account, and OpenClaw's agent turns then count against your plan's Codex allowance instead of an API bill. OpenAI lists OpenClaw among its open-source Sign in with ChatGPT integrations, and OpenClaw's own docs say "OpenAI explicitly supports subscription OAuth usage in external tools and workflows like OpenClaw."

The catch is choosing the right login and model. OpenClaw now offers three ways to connect OpenAI, they do not all support the same features, and the default subscription model is the one that drains a Plus plan fastest. We read OpenClaw's provider docs and OpenAI's pages on 11 October 2026; OpenClaw's latest release that day was v2026.10.1.

How do you connect OpenClaw to your ChatGPT plan?

OpenClaw's OpenAI authentication page gives one command per method. Run it on the machine where OpenClaw is installed:

# Codex login through a browser (the default)
openclaw models auth login --provider openai --method oauth

# Codex login on a remote or headless machine
openclaw models auth login --provider openai --method device-code

# Sign in with ChatGPT (Beta)
openclaw models auth login --provider openai --method siwc

Or use the dashboard: Settings, Models, select the agent, Connect provider, OpenAI, then pick a sign-in method. For Sign in with ChatGPT, OpenClaw says you must "approve token sharing in your browser"; if you sign in with identity permissions only, the account saves but "cannot authorize inference."

Then check it worked:

openclaw models list --provider openai
openclaw models status

In chat, /status shows the active runtime and /codex status confirms the native Codex runtime. One change from older guides: the setup page says onboarding "no longer imports OAuth material from ~/.codex", so sign in through OpenClaw even if the Codex CLI is already logged in. Older codex/* and openai-codex/* model names are repaired by openclaw doctor --fix.

Codex login or Sign in with ChatGPT: which should you pick?

Both use "your Codex allowance", according to OpenClaw's comparison table. They differ in what else works.

	Codex login (browser or device code)	Sign in with ChatGPT (Beta)
Billing	Your Codex allowance	Your Codex allowance
How OpenClaw is identified	You, using the Codex product	OpenClaw as its own authorised app
OpenClaw tools and local plugins	Supported	Supported
OpenClaw image generation	Supported with account access	Not supported by this credential
Audio transcription and realtime voice	Supported on some routes	Not supported by this credential
OpenAI-hosted plugins	Needs a connector-scoped grant	Not supported yet
Permission control	Codex product permissions and workspace policy	App-specific OAuth grant, managed in ChatGPT login connections

For most people running a personal agent, the default Codex login is the safer choice today because it supports more features. Pick Sign in with ChatGPT if you want OpenClaw to appear as its own connected app you can revoke on its own. Note that OpenClaw "does not set per-app SIWC limits or show per-app usage", so check usage in ChatGPT Settings, Usage, where OpenAI lets you cap each connected app at a percentage of your weekly usage.

The third method, an OpenAI Platform API key (--method api-key), bills per token. OpenClaw's setup page shows how to list a subscription profile first and an API key second under auth.order.openai, so the agent falls back to the key when the plan runs out. Only do that if you are happy to pay API rates past your limit.

Which model should you run OpenClaw on with a ChatGPT plan?

OpenClaw's fresh subscription setup selects openai/gpt-6-astra. That is a capable model, and also the most expensive against your plan. OpenAI's Codex pricing page estimates how many local messages Plus covers in one five-hour window:

Model	Plus, per five hours (OpenAI estimate)
GPT-6 Astra	5 to 45
GPT-6.1 Sol	15 to 160
GPT-6 Sol	15 to 150
GPT-6 Luna	350 to 3,000

OpenAI adds that weekly limits may also apply, that "Pro plans currently have no five-hour limit", and that Fast mode uses included limits at 2.5 times the Standard rate. An always-on assistant answering chat messages, running scheduled tasks and calling tools makes many turns you never see, so on Plus it is worth moving routine work to Luna or Sol:

openclaw models list --provider openai
openclaw config set agents.defaults.model.primary openai/<model-from-that-list>

Use the exact refs the list shows for your account; OpenClaw's docs say the subscription catalogue follows what your signed-in account exposes. Also note that GPT-5.5 leaves ChatGPT and Codex on 14 October, so any agent still pinned to openai/gpt-5.5 on a subscription needs a new model. Our post on the GPT-5.5 retirement has the details.

Which ChatGPT plan works with OpenClaw?

Plan for Plus or Pro. OpenAI's Sign in with ChatGPT page says "eligible ChatGPT Plus and Pro subscribers" can use their plan in supported apps, and its pricing page lists Free and Go access as "GPT-6 Luna at Standard speed in the desktop app." "OpenClaw chatgpt go" is a common search, but we did not find an OpenClaw or OpenAI page that says a Go plan works through OpenClaw, so treat it as unconfirmed.

What problems are OpenClaw users hitting with ChatGPT sign-in?

Three open issues on the OpenClaw repository are worth knowing about, all still open on 11 October:

• Sign in with ChatGPT consent error. Issue #161216, opened 29 September, reports the beta consent page returning invalid_client ("This app is unavailable"). The reporter's workaround was a different credential. If you hit it, use Codex login instead.
• False "Credentials rejected". Issue #145086 reports the Control UI marking a working ChatGPT subscription login as rejected and the picker showing "sign-in needed", even though the model still runs through the Codex harness. Test with a real message before re-authenticating.
• Usage-limit error without a reset time. Issue #168288, opened 10 October, reports OpenClaw returning a Codex usage-limit error on a minimal request while a native Codex test on the same Pro account succeeded. The reporter notes the two tests used different auth state, so it is not yet a confirmed bug.

When something breaks, OpenClaw's setup page suggests openclaw models auth list --provider openai, then signing in again and running openclaw models status --probe --probe-provider openai.

Can OpenClaw use a Claude subscription instead?

Yes, but only one way, and it works differently from ChatGPT. OpenClaw's Anthropic page offers a "Claude CLI" route that reuses "an existing Claude Code login through the installed executable on the same host." You install Claude Code, log in with claude auth login, then choose Claude CLI in openclaw onboard. OpenClaw says it "never reads, persists, refreshes, selects, or forwards the native login tokens."

That matters because of Anthropic's rules. Anthropic's Claude Code legal page lets you sign in to the unmodified Claude Code with your own subscription, including when another platform runs it, but does not allow developers "to route requests through Free, Pro, or Max plan credentials on behalf of their users." In April 2026 Anthropic stopped covering OpenClaw's older direct Claude login with plan limits, as VentureBeat reported. The Claude CLI route runs Anthropic's own binary, which is why it is the one we list. Skip community proxies that turn a Claude plan into an API endpoint.

Two practical differences from the ChatGPT side: Claude has no per-app cap, and OpenClaw does not poll quota for native Claude CLI logins, so watch Settings, Usage on claude.ai yourself. Our guide Can I use my Claude subscription in other apps? explains the rule in full.

Other agents that work the same way

If OpenClaw is more than you need, several listed apps also run on a ChatGPT plan: Hermes Agent (also a Sign in with ChatGPT partner), OpenCode (see our OpenCode setup post), Captain Claw and Clawboo. The ChatGPT hub lists them all, the OpenClaw listing has the short version of this post, and our open-source agents guide compares OpenClaw with the other open-source options.