At DevDay 2026 on September 29, OpenAI announced an expansion of "Sign in with ChatGPT," which lets people use the AI allowance included in ChatGPT Plus and Pro in supported third-party apps.
A wide range of users can log in with a ChatGPT account. But only eligible Plus and Pro subscribers can use their ChatGPT plan allowance for AI processing in external apps. In supported tools such as Devin, Notion, and Vercel, users who grant permission can use the allowance included in ChatGPT Work and Codex for AI tasks.
Because users no longer need to create their own OpenAI API key or set up separate API billing, they can keep the ChatGPT plan they already pay for and more easily choose the outside app that suits their needs.
That said, a monthly ChatGPT fee does not unlock every feature. Apps have their own pricing, and some processing falls outside the ChatGPT plan. This arrangement separates AI model allowance from the cost of the app that uses it, and its effect on user choice and developers' business models is drawing attention.
Logging in with ChatGPT is not the same as using your allowance
Signing in to other services with a ChatGPT account already has precedents.
On July 29, Supabase announced a beta authentication feature that lets people sign up or link existing accounts with a ChatGPT account.
Supabase's earlier announcement
What OpenAI expanded at DevDay goes beyond login: supported tools can now also use the AI allowance included in a ChatGPT plan.
When a user chooses "Continue with ChatGPT" or "Sign in with ChatGPT," they authenticate on the ChatGPT side and review the basic information that will be shared with the external app.
The shared information includes name, email address, and, if one is set, a profile picture.
However, logging in to an external app and allowing that app to use the ChatGPT plan allowance are authorized separately.
A successful login does not mean the app can automatically use the ChatGPT allowance.
In addition to being a Plus or Pro subscriber, the app must support plan usage, and the app itself must have selected eligible plans or features.
As of October 2, 2026, OpenAI's list of supported services divides them as follows:
| Category | Services and tools listed in official guidance | ChatGPT plan allowance |
|---|---|---|
| Commercial tools that support plan usage | Amp Code, Conductor, Dactyl, Devin, Hermes Agent, Hyperagent, Kilo Code, Notion, Vercel, Vorflux, Warp | Eligible Plus and Pro users can grant permission to use it |
| Open-source integration examples | OpenClaw, OpenCode, Pi, T3 | Eligible plan allowance can be used from supported tools |
| Planned | Lovable | Listed as "coming soon" |
| Sign-in only | Airtable, Canva, GitLab, HubSpot, Supabase | Not currently eligible for plan usage |
OpenAI mentioned "16 partners" at DevDay, but users will need to check each one's availability before relying on it.
Even for services like Supabase that allow login with a ChatGPT account, the AI allowance of a ChatGPT plan may not be usable.
The list of supported services may grow, but plan usage for commercial services is currently offered mainly through partners selected by OpenAI.
Also, allowing an external app to use your ChatGPT plan does not mean the app can automatically read your ChatGPT conversation history or memory.
Text and files entered into an external app are also subject to that app's own features and privacy policy.
Letting an app use your ChatGPT plan allowance and sharing personal information or conversation history inside ChatGPT are handled as separate mechanisms.
What the monthly ChatGPT fee covers, and what costs extra
Vercel's app-building service v0 shows concretely where the boundary lies.
According to Vercel, processing by supported OpenAI models can use the ChatGPT plan allowance, while some operations, such as image generation and delegated sub-agents, may consume v0's own credits.
When the ChatGPT allowance runs out, processing stops and progress is preserved, and the user can choose whether to continue using v0 credits.
What a ChatGPT plan covers is the allowance for supported AI processing. It does not mean every feature or charge of an external app is included in a ChatGPT subscription.
Based on public information from OpenAI and Vercel as of October 2, 2026, the breakdown by type of processing or cost looks like this:
| Item | Relationship to ChatGPT plan | Conditions and costs to check separately | Source |
|---|---|---|---|
| Login to external apps | Authenticates using ChatGPT identity | Plan usage requires separate permission | Quickstart |
| Supported AI processing | Consumes the ChatGPT Work/Codex allowance included in Plus and Pro | Supported app, eligible user, plans supported by the app | OpenAI usage guide |
| App paid features and execution infrastructure | Not bundled into the ChatGPT plan | App's own subscription fees, infrastructure, add-on features, etc. | OpenAI usage guide |
| v0 image generation, sub-agents, etc. | Separate from the portion processed under the ChatGPT plan | May consume v0 credits | Vercel's guide |
| Responses API route for open source | ChatGPT plan can be used for supported AI processing | Image generation, hosted file search, Code Interpreter, and others are not supported | Preview limitations |
| After the ChatGPT plan allowance is used up | ChatGPT credits can be used if conditions are met | Requires explicit permission, a balance, per-app limits, etc. | OpenAI usage guide |
The distinction to keep in mind is between the cost of running the AI model and the cost of the app performing the whole task.
For example, assuming that code-execution environments or an app's own features are covered by the monthly ChatGPT fee could lead to a mismatch with actual charges.
OpenAI also advises developers to display what is covered by the ChatGPT plan separately from the app's own fees.
On September 29, Warp also announced that Warp Terminal and Warp Agent CLI can now use the ChatGPT Work and Codex allowance for AI processing.
The same ChatGPT plan can thus be used for app development in v0 and for terminal work in Warp.
The contract that pays for AI usage and the app where the work actually happens no longer have to be the same service.
Using external apps doesn't increase the allowance itself
Usage in external apps is drawn from the same allowance shared with ChatGPT Work and Codex.
Connecting an app does not add a new allowance.
For example, if you prototype an app in v0 during the day and fix code in Codex at night, both draw on the shared ChatGPT plan allowance.
In ChatGPT's "Settings > Usage," you can set a cap on how much each app can use.
However, this cap is a limit to keep an app from overusing the shared allowance. It does not reserve usage in advance.
If another app or Codex exhausts the shared allowance, an app may be unable to use the ChatGPT plan allowance even if it hasn't hit its own cap.
OpenAI's explanation of usage limits
While it becomes easier to try multiple apps, users also need to be mindful of how much allowance they spend on each task.
Rewriting a short passage and handing an agent a long-running task will not necessarily consume the same amount.
OpenAI also explains that how quickly the allowance is used up differs by app and by the work performed.
For that reason, the fact that you can log in with ChatGPT alone doesn't tell you how much of your regular work can be covered by the monthly fee.
The feature that uses ChatGPT credits after the plan allowance runs out is disabled by default.
Supported accounts must explicitly allow it, and the app's usage limit must be set to 100%.
However, setting an app's cap to 100% alone does not enable credit usage. Other conditions also apply, such as the account being eligible and having an available credit balance.
In addition, the setting that allows use of credits you hold is separate from the setting that automatically purchases credits.
If auto-purchase is enabled and you keep using the app, additional charges may be incurred.
OpenAI's credit usage conditions
You should also be careful with connections to apps you no longer use.
Simply logging out of an external app does not revoke its connection to ChatGPT or its permission to use your plan.
To fully disconnect, remove the app from "Settings > Security and login > Sign in with ChatGPT" in ChatGPT.
If you want information already given to the external app deleted, you need to go through that app separately.
Even small developers can lower the barrier to getting started
"Paste Perfect," a sample implementation released by OpenAI, is an open-source desktop app that rewrites copied text before pasting it into another app.
Previously, users would have had to create their own OpenAI API key and even set up API billing.
With Sign in with ChatGPT, eligible Plus and Pro users can log in with their ChatGPT account and rewrite text using the allowance of the plan they already have.
OpenAI's sample implementation
For small, narrowly focused apps, setup before first use can hinder adoption.
If someone has to understand API pricing, manage an API key, and set up a separate payment method just to try a text-rewriting tool, some will give up at that point.
If the existing ChatGPT plan can be used as is, developers can focus on usability and workflow design, and users can try the app before setting up any additional billing.
However, this is a benefit expected from how the system works, not evidence that the feature has actually increased user numbers or by how much it has cut costs.
For developers, an OAuth mechanism is provided that handles authentication and authorization through the browser.
In the open-source route, a client is registered at first login, and the app checks whether the user has allowed use of their ChatGPT plan.
Neither a secret client key nor the user's own OpenAI API key is required.
On the other hand, the access and refresh tokens the app receives are credentials and must be stored securely.
For AI processing, the app calls OpenAI's public Responses API using the access token obtained through OAuth.
First, it retrieves the list of models available to the signed-in ChatGPT account, then specifies one of them.
When using the Responses API over HTTP, store: false and stream: true must be set.
This is not a method of calling ChatGPT's internal private APIs, but an official route documented by OpenAI for external apps.
Also, an allowance-related error can be returned even after a streamed response has begun. An app should therefore not treat processing as successful just because part of the text arrived, and should confirm that it received response.completed at the end.
Model and inference specifications
There are also limits on available features.
In the current open-source preview, some Responses API features are unavailable, including image generation, hosted file search, Code Interpreter, and Computer Use that runs on OpenAI's side.
On the other hand, combining it with the Codex app-server makes it possible to run shell commands, MCP tools, and the like in a local environment.
In other words, having permission to call an AI model doesn't mean OpenAI provides all the execution environments and tools an app needs.
There are conditions on distribution as well.
The developer mechanism OpenAI has published so far targets open-source apps and personal projects that run locally.
Developers who want to offer plan usage in paid apps or remotely hosted commercial services must apply to OpenAI to take part.
Eligibility and execution environments
A method for running open-source apps on a user's own virtual machine has also been published.
However, for sessions to which credentials have been forwarded, it currently does not support tracking usage per execution host or revoking plan-usage permission for that host alone.
The release of a mechanism for local apps does not mean any commercial cloud service can freely use a ChatGPT plan.
The ChatGPT plan can be separated from the app you use
The big change for users is that they can keep their ChatGPT allowance while choosing which app to actually work in.
People who want to develop in a terminal, use a dedicated text-rewriting tool, or prototype a web app can each pick a supported app suited to their purpose while using the same ChatGPT plan.
This does not make every feature of external apps free, but it reduces the hassle of setting up an OpenAI API key for each app or signing a separate contract for AI model usage.
For app makers, it opens room to price the value specific to the app, on the premise that users bring their own AI model allowance.
Examples include mechanisms for handing materials to the AI properly, environments for running tools safely, and interfaces that let users check progress on long-running work and resume partway through.
Even when the same OpenAI models are available, how efficiently real work gets done depends on the app's design.
As more apps support ChatGPT plans, users may find it easier to choose apps by comparing these usability and surrounding features, not just the model itself.
On the other hand, because multiple apps share the same ChatGPT plan allowance, its management and terms of use depend on OpenAI's mechanisms.
While switching between apps becomes easier, changing or canceling a ChatGPT plan could affect multiple supported tools.
A shared allowance that is convenient for users could also become one more reason for them to keep their ChatGPT plans.
However, this is a business possibility inferred from the service's structure, not an effect demonstrated by usage rates or subscription retention rates.
When choosing a supported app, it's worth checking whether it offers the models and features you need, how much of your ChatGPT plan allowance it consumes, and what separate fees the app itself charges.
If more services present these conditions clearly, users will be able to treat the ChatGPT plan as a common AI foundation while choosing the most usable app for each job.