ChatGPT Business plans now work in OpenCode, Warp and Notion, two days before Anthropic bans routing Claude through consumer subscriptions

OpenAI now lets ChatGPT Business workspaces spend their plan in other companies' AI tools. "Now, you can use your ChatGPT Business plan in tools like Notion, v0, Amp, Kilo Code, Warp, Hermes from Nous Research, OpenClaw, OpenCode, and more," OpenAI Developers posted on November 10, pointing to Sign in with ChatGPT. Before this, the feature was pitched at personal plans; OpenAI's developer guide still speaks of "eligible ChatGPT Plus and Pro users."

The timing is the interesting part. Two days later, on November 12, Anthropic's rewritten Usage Policy takes effect, and its "Do Not Abuse Our Platform" section gains a new line against services that "resell, proxy, or otherwise provide access to Claude through unauthorized means, including services that route requests through consumer subscriptions or misrepresent the product or client being used." One lab is turning its subscription into a currency other apps can accept. The other is writing down that its own may not be spent that way.

What Business admins actually get

OpenAI's Help Center article and its ChatGPT Learn page describe two separate switches, both in the admin console:

• Per app. Under Tenant settings → External access, an admin picks an app and sets "Use ChatGPT plan" to Allowed. Nothing is allowed until someone does this app by app.
• Per member. Under the workspace's Usage limits → ChatGPT plan, "Usage in other services" sets a percentage of each Standard or Premium member's included allowance that all approved services together may use. It is a ceiling, not a separate budget: OpenAI says it "does not create a separate pool of usage for each app." Members who hit it can move on to workspace credits if they have permission; usage-based members only ever use credits.

Members can then set tighter caps of their own in Settings → Usage, but can't loosen the admin's. Signing in and spending the plan are separate permissions throughout, and OpenAI repeats that plan sharing gives the app no access to ChatGPT conversations or memories.

What gets spent is the "ChatGPT Work and Codex allowance." That detail will matter more in a few weeks: from December, as we reported on October 31, ChatGPT and Codex will share one token-based pool, so an agent running in Warp or OpenCode will be drawing on the same allowance as the user's own chats and Codex tasks.

The partner list has one asterisk worth knowing about: Devin supports ChatGPT plan usage for personal plans but "does not support ChatGPT plan usage for ChatGPT Business plans." OpenAI's post also names v0, while the directory lists Vercel.

How the plumbing works

The developer side is documented in OpenAI's cookbook guide for open-source apps, and it is ordinary OAuth with one unusual piece. A desktop app opens the browser with PKCE and a localhost callback, sends client_id=dynamic_agent_client and an app-generated host ID, and OpenAI registers a new client during the flow, so the app needs no client secret and no API key. The permission that matters is the scope chatgpt.tokens.use.direct on the resource https://api.openai.com/v1. With it, the app calls the public Responses API using the user's OAuth access token, and the cost lands on the user's plan.

OpenAI is not opening this to everyone. Plan usage is available to "open-source projects, personal projects that run locally, and selected private apps"; paid or hosted apps have to join a waitlist. The feature itself is not new either. Codex lead Thibault Sottiaux (Tibo) was already pointing people to a thread on how the subscription works in partner products on October 1, and on October 5 he said a default speed-up for GPT-6 Astra and GPT-6.1 Sol applied "across all our products and partners using Sign in With ChatGPT (including OpenCode, Pi, Amp, Devin, ...)". What changed on November 10 is the step into company accounts, where an admin decides.

Anthropic's side of the same week

Anthropic's position has been consistent, and from November 12 it is also in the Usage Policy. Its Claude Code legal and compliance page already says that OAuth sign-in is for "ordinary use of Claude Code and other native Anthropic applications," that developers using the Agent SDK "should use API key authentication," and that Anthropic "does not permit third-party developers to offer Claude.ai login into their own applications, or to route requests through Free, Pro, Max, or Ultra plan credentials on behalf of their users." The new policy line puts that into the document that applies to every user and comes with the usual penalties: warnings, throttling, suspension or termination. Anthropic's October 8 post about the update focused on deceptive campaigns, weapons, surveillance and high-risk uses, and didn't mention subscriptions.

OpenCode, on both lists, shows how this plays out. In March it merged a pull request titled "anthropic legal requests" that removed its built-in Anthropic auth plugin. On October 30 it merged "feat(core): add Claude Agent provider", which takes a different route: the user installs Anthropic's own Claude Code, runs claude auth login, and OpenCode drives it through the Claude Agent SDK, so OpenCode never touches the Claude credentials. When a user asked on the PR whether Anthropic had approved this, the author, Tarquinen, an OpenCode organization member, replied two minutes later: "We got permission from Anthropic and everything here is fully tos compliant." We found no public confirmation from Anthropic: nothing on its X accounts since late October, and nothing on the legal page or in the Agent SDK docs naming OpenCode. Users were still asking for proof on November 8 and 9.

Whether that route falls under the new line is Anthropic's call. The legal page explicitly allows an end user "signing in to the unmodified Claude Code binary with their own Claude subscription," which is what the Claude Agent provider does; the same paragraph tells Agent SDK developers to use API keys. The policy line covers "consumer subscriptions," and Team and Enterprise are commercial plans under Anthropic's Commercial Terms, so the line as written targets personal plans, not the kind of company workspace OpenAI just opened up.

What it means if you pay for both

For a team on ChatGPT Business, the path is now official and admin-controlled: allow the specific apps, pick a percentage, and expect that percentage to come out of the same pool as Codex and, from December, Chat. For someone on Claude Pro or Max, the safe options in a third-party agent are still an API key, a cloud provider, or Anthropic's own Claude Code. The OpenCode route depends on a permission that, publicly, exists only in one GitHub comment, and the risk of a wrong guess sits with the user's account, not with OpenCode.