Sign in with ChatGPT: the user's plan pays your bill

Every AI product built in the last two years carries the same wound: you pay for the tokens, the customer pays you a flat monthly fee, and the heavy users eat the margin. At DevDay 2026 on September 29, OpenAI offered a way out that is also a leash. Sign in with ChatGPT lets Plus and Pro subscribers spend their own OpenAI allowance inside your product.

What actually happened

Sixteen partner tools went live with it, including Devin, Notion, Vercel, Warp, Amp, OpenCode, OpenClaw, Kilo Code, Conductor, T3, and Dactyl, per The New Stack, with Lovable listed as coming soon. A Plus or Pro user signs in with their ChatGPT account and eligible model usage inside that app draws down their OpenAI plan instead of the vendor's API bill.

Two controls sit on top. Users set a weekly cap per app in ChatGPT settings, and OpenAI's help documentation confirms that when the cap is hit, included usage simply stops — it does not silently roll over to the partner's own billing. Settings also breaks down how much of a user's ChatGPT usage came from each connected app. On the identity side, the sign-in shares name, email, and profile picture only; conversations, memory, files, and billing details stay on OpenAI's side.

Why subscription portability matters for your business

Your unit economics just became someone else's variable. If your power users route their inference through their own ChatGPT plan, your COGS per seat drops toward zero. That is real. It also means OpenAI now sets the ceiling on how much product your customer can consume, and OpenAI moved that ceiling twice today — Pro 200's Codex and Work allowance is being halved on October 30. When their quota tightens, your product feels broken and you did nothing.

Build for a hard stop, not a soft degrade. Usage stops at the cap. Your app needs to detect that state and handle it like a first-class path: tell the user their weekly cap is spent, offer to fall back to your own billed capacity, and keep their work recoverable. A spinner that never resolves is how a quota ceiling becomes a churn event.

Keep your own key path alive. The pattern that survives is two-track auth: identity from wherever the customer prefers, inference through an interface that can point at their ChatGPT allowance, your API key, or another provider entirely. Make it a routing decision logged per request. Products that hard-wire themselves to a single subscription rail inherit every pricing change that rail makes — and there were two of those in one keynote.

Sources: The New Stack, OpenAI Help Center.