If you’ve been running OpenClaw with a Claude Pro or Max subscription, you’ve probably seen the news. Anthropic updated (or clarified) their terms of service, and the new language is pretty clear:
OAuth authentication (used with Free, Pro, and Max plans) is intended exclusively for Claude Code and Claude.ai. Using OAuth tokens obtained through Claude Free, Pro, or Max accounts in any other product, tool, or service — including the Agent SDK — is not permitted and constitutes a violation of the Consumer Terms of Service.
The frustrating part: nobody at Anthropic will confirm whether this includes OpenClaw, or better yet, what this really means. People are asking. Very vague to no straight answers are given.
Read that paragraph again though. OpenClaw uses OAuth tokens from Claude Pro/Max accounts. OpenClaw is not Claude Code. It’s not Claude.ai. It’s a third-party tool. You can connect the dots.
Most of the OpenClaw community has connected those same dots and are quickly planning a move to ChatGPT’s Codex subscription. OpenAI doesn’t have a problem with their OAuth tokens in third-party tools. So that’s where I’m going too.
Here’s how to make the switch.
It takes about 5 minutes.
Before You Start
You need an active ChatGPT subscription (Plus, Pro, or Team). The Codex model (gpt-5.3-codex) is available through this subscription.
Your OpenClaw config, workspace, memory, skills, cron jobs, all of that stays. You’re only swapping the model behind it.
However, DO make a backup of your .openclaw folder before doing this. It’s always good practice.
Step 1: Run the Onboard Command
openclaw onboard --auth-choice openai-codexThis starts the onboarding wizard. A few things to watch for:
When it warns you this is risky: Acknowledge it and continue. It’s just telling you that switching providers might affect behavior since different models have different strengths.
When it asks about setup mode: Choose Quickstart.
When it asks about your existing values: This is the important one.
⚠️ Choose “Use existing values.” Not “Update values.” Not “Reset.”
If you choose Reset, it wipes your config. Your channels, your memory settings, your cron jobs — gone. Don’t do that. “Use existing values” keeps everything intact and only adds the new OpenAI auth.
Step 2: Authenticate With ChatGPT
The wizard opens ChatGPT in your browser. Sign in with your ChatGPT account and authorize the connection.
Once you’ve authenticated, you’ll be sent back to the terminal.
Step 3: Stop Here (Don’t Continue Setup)
After the browser auth completes, the onboard wizard will try to continue to the next step, which is setting up channels. You don’t need to do this. Your channels are already configured.
Just close the terminal (Ctrl+C or close the window). The auth token is already saved.
Step 4: Set the Model
Now tell OpenClaw to use the Codex model:
openclaw models set openai-codex/gpt-5.3-codexStep 5: Verify
Confirm it worked:
openclaw models status --plainYou should see openai-codex/gpt-5.3-codex as your active model.
Send a test message to your agent through whatever channel you use (Discord, Telegram, etc.) and make sure it responds. If it does, you’re done.
What About API Keys Instead?
You also have the option of using an OpenAI API key (pay-per-token) instead of the ChatGPT subscription OAuth:
openclaw onboard --auth-choice openai-api-keyThis uses usage-based billing through the OpenAI Platform instead of your subscription. Whether that’s cheaper depends on how much you use OpenClaw. For heavy users, the subscription is probably the better deal.
Will GPT-5.3-Codex Be As Good As Claude?
Honestly, I’m still figuring this out. Claude was excellent at following complex instructions and writing that sounded natural. Codex is strong at code and tool use.
You may need to tweak your SOUL.md and AGENTS.md. Different models interpret the same prompts differently. Give it a few days before you judge it or make any drastic moves.
What About Staying on Claude?
You have two options if you want to keep using Claude:
Anthropic API key (not OAuth) — Go to console.anthropic.com, create an API key, set up billing. This is pay-per-token and Anthropic explicitly says this is the approved method for developers. Run:
openclaw onboard --auth-choice anthropicWait for clarification — Anthropic might eventually clarify whether OpenClaw specifically is affected. But they’ve had weeks and haven’t, so don’t hold your breath.
The API key route keeps you on Claude legally, but you lose the flat subscription pricing and go to per-token billing. Depending on your usage, that could be cheaper or a lot more expensive.
Wrapping Up
The whole switch takes about 5 minutes. Your workspace, memory, cron jobs, and channels all stay intact. The only thing that changes is which model answers.
# The full sequence
openclaw onboard --auth-choice openai-codex
# → Choose Quickstart
# → Choose "Use existing values"
# → Authenticate in browser
# → Close terminal after auth (don't continue setup)
openclaw models set openai-codex/gpt-5.3-codex
openclaw models status --plainWould be nice if Anthropic just said “yes, this includes OpenClaw” instead of letting everyone guess.
If they do, I will update this post ASAP.
But enough people have read the same TOS and landed on the same conclusion. Better to switch now than wake up to a suspended account.
What do you think? 👇
This page may contain affiliate links. Please see my affiliate disclaimer for more info.
