guide

Who pays for AI tokens? BYOK, ChatGPT or AI Pass

ChatGPT plan usage removes an inference cost. AI Pass lets developers add a markup on eligible paid usage. Compare both with app-funded APIs and BYOK, including limits and setup.

EiliyaSeptember 30, 20267 min read

Your app can stop paying for eligible AI requests without earning anything from them. That's the distinction behind who pays for AI tokens: Sign in with ChatGPT uses the user's existing plan, while AI Pass uses a wallet that lets you add a developer markup.

A subscription can pay for your software while inference comes from your API account, the user's provider key, a ChatGPT allowance, or a portable wallet such as AI Pass. Each route assigns the bill and the responsibility differently. OpenAI's quickstart separates permission to use the plan from identity sign-in, so a login alone doesn't settle who pays.

Who pays for AI tokens in each model?

Start with the inference bill. The margin effects below are our analysis of who carries that cost; model quality is a separate comparison.

Funding route	Who funds inference?	Effect on app margins	User setup	Main trade-off
Developer-owned API key	The app's provider account	More usage can reduce contribution margin unless pricing covers it	Ordinary app onboarding	The app carries usage risk and controls the provider choice
Bring your own key, or BYOK	The user's provider account	The app avoids that provider bill	Create a provider account, arrange billing, and configure a key	The user must understand and trust credential handling
Sign in with ChatGPT	Included plan usage, or permitted ChatGPT credits	Cuts the app's inference cost; the app earns nothing on that usage	Authorize the participating app	OpenAI eligibility, shared limits, and preview scope apply
AI Pass wallet	The user's AI Pass balance	A developer markup can turn paid usage into income	Connect and fund a wallet	Usage remains payable and adds a gateway dependency

OpenAI's billing guide explains the provider-key distinction. Its plan-usage documentation covers the ChatGPT route, while AI Pass's integration documentation covers the wallet.

Your API key puts inference on your bill

When you own the API account, you can offer one checkout and include inference in your price. Users don't have to know which provider handled the request. For a customer who's never created an API key, that's a useful way to start.

Your price still has to survive actual usage. A long agent session can mean repeated model requests, tool results, and retries; if the customer's app fee stays flat, you cover the extra resources.

Ask what work your price includes, what you meter, and what happens at the limit. A flat software fee with explicit usage allowances is easier to defend than unlimited agent work your provider bill can't support.

OpenAI says API charges are separate from ChatGPT subscriptions and can include tools or storage as well as tokens. Use your own provider key and you still get that bill, even when your customer subscribes to ChatGPT.

BYOK moves the bill and leaves the key handling

With BYOK, the user brings a provider credential. A provider-direct design bills model usage to their provider account, leaving you free to charge for the workflow without bundling inference into the price.

That suits technical users who already manage API accounts and want a direct provider relationship. Someone who just wants to rewrite a document has less reason to learn about provider projects, keys, and separate billing dashboards.

Explain where the key lives: does it stay on the user's device or go to your server? Who can use it, and how can the user revoke it? The BYOK label answers none of those questions.

In a February 2026 developer-forum discussion, aprendendo.next raised key custody and users' reluctance to create developer credentials. The proposed alternative was provider account authorization. It's one discussion rather than a survey, but the friction is familiar: eligible OpenAI requests can now use a ChatGPT sign-in flow instead.

ChatGPT plan usage cuts costs without inference earnings

A subscriber can authorize ChatGPT plan use without opening a separate API billing account. Eligible requests within included usage need no separate inference purchase, so your app's AI is effectively free to that user within their paid allowance. You earn nothing on that inference, though you can still charge for your product.

OpenAI reports 1.2 billion weekly ChatGPT users, a compelling audience for an app developer. That's its reported audience, not the number of subscribers eligible for every integration. The partner directory includes Notion, Devin, Vercel, and Kilo Code.

The user buys capacity through a subscription and shares that allowance across work. An app-specific cap limits what one app can consume; it doesn't create another pool. Heavy users stop adding to your inference bill, but shared limits can still interrupt their tasks.

OpenAI's Help Center describes per-app weekly caps and separately authorized credit use. Buying capacity and allowing an app to spend it are separate choices. Keep the paid allowance and its limits visible when you describe included AI usage.

You also need access to offer it. The public overview directs paid and remotely hosted apps to an interest form. The preview has a defined Responses API contract that excludes image generation and hosted connectors, among other capabilities; check whether your product fits before choosing the funding route.

AI Pass lets paid usage become developer income

With AI Pass, the user pays for exact model usage from one portable wallet across participating apps. You can add a markup and earn on eligible paid usage under the current AI Pass terms. Instead of just removing an inference cost, usage can become income without you funding every call, selling another subscription, or asking for provider keys.

No user subscription is required, and AI Pass covers text, images, speech, and video across multiple providers. You can integrate through its JavaScript SDK or OAuth plus an OpenAI-compatible REST API. AI Pass is independent of OpenAI, with no affiliation or endorsement.

The developer route is open today without an application form. That gives you a way to build user-funded usage now, rather than making commercial access depend on OpenAI selecting your app.

OpenAI has the stronger starting point for users who already pay for an eligible plan and want supported OpenAI models. Included usage needs no wallet top-up, and per-app caps stay in ChatGPT. AI Pass suits users who want multiple providers or usage without a subscription, at the cost of another balance to manage and gateway to trust.

Heavy use still costs money; the wallet charges it to the user's balance. Metering can suit occasional use. If you run agents all day, compare the actual workload and available allowance before assuming either route is cheaper.

Try it today: ask your agent

Open AI Pass's docs, pick your tool, and paste the one brief into Claude Code, Cursor, Codex, Lovable, Replit, or another capable coding agent. Your agent inspects the project, chooses a working integration path, and sets it up with your approval in the browser.

There's no application form, waitlist, provider-key handoff, or manual OAuth client setup. You sign in to AI Pass to approve the setup. Commercial or hosted Sign in with ChatGPT apps, by comparison, go through OpenAI's interest form and partner selection.

Tell users what your fee buys and which balance pays

If customers pay you for a workflow, collaboration, or hosting, explain that software charge separately from model funding. A model picker shouldn't double as a billing puzzle.

Show the active funding route where the user makes a request. A ChatGPT connection, provider key, and wallet balance grant different permissions with different consequences. OpenAI's UI guidance asks apps to distinguish their charges from plan usage and provide a usage-management action.

I'd choose the route that fits your audience and make alternatives explicit. Consumer onboarding benefits from fewer credential steps, while technical users may prefer direct provider billing. If your product uses multiple providers, its funding path needs to cover those models.

When funding fails, preserve the task, explain the limit, and let the user choose another route. Renewing a login won't replenish a spent allowance.

FAQ

Does ChatGPT Plus include ordinary API-key usage?

No. OpenAI Platform billing is separate. Authorized plan usage in a participating app is a distinct integration.

Is BYOK the same as Sign in with ChatGPT?

No. Conventional BYOK supplies a provider API key, while Sign in with ChatGPT uses account authorization and eligible plan usage. Some apps use BYOK as a broader label in their settings.

Can an app charge while using my ChatGPT plan?

Yes. Its software or service charges can be separate from inference funded by your plan.

Does a user-funded wallet require a subscription?

AI Pass does not. Users pay for exact usage from their balance. The host app's own charges still need to be explained separately.