Skip to content
Summer sale. 50% off Pro, $19.50 for life. Desktop app: Apple Silicon Mac only.SUMMER50
CodexUse

Is running a custom model provider in Codex allowed?

PolicyUpdated September 2026 · 4 min read

Here is what a custom-model account does mechanically, what it does not do, and which terms are yours to check.

Short answer: Codex supports custom model providers in its own documented configuration, and the app ships first-class non-OpenAI authentication modes. A custom-model account uses that supported path with your own provider credential. It is not a way around OpenAI billing, because it does not use OpenAI’s models. You are responsible for your provider’s terms and your own plan terms.

What it mechanically does

  • Creates an isolated Codex home — the same mechanism the Codex CLI documents for separating accounts.
  • Sets a custom model provider in that account’s configuration. Codex reads a model_providers table by design; this is not a patch or an injection.
  • Authenticates that account with a locally minted credential rather than a ChatGPT login. The app already supports non-ChatGPT authentication modes, including API-key and Amazon Bedrock accounts.
  • Routes that account’s model requests to the endpoint you configured, with the credential you supplied.

What it does not do

  • It does not use OpenAI’s models, quota, or inference on that account. Nothing is consumed on OpenAI’s side to serve those turns.
  • It does not modify, patch, repackage, or redistribute the Codex app. The app is the one you installed, configured through its own settings.
  • It does not create accounts, forge identities, or share one subscription between people.
  • It does not bypass a paywall. A custom-model account gets no ChatGPT features — hosted features are unavailable there precisely because there is no ChatGPT account behind it.
  • It does not give you OpenAI models for free. If you want those, use an OpenAI account and your own account.

What is yours to check

  • Your provider’s terms. You are their customer and you pay them directly. Rate limits, acceptable use, and data handling are between you and them.
  • Your own OpenAI plan terms, if you also run OpenAI accounts. Those are unchanged by this feature, and our multiple-accounts guide covers that side.
  • Your employer’s policy, if the work is theirs. Some organizations require a specific gateway; some prohibit third-party endpoints entirely.
  • Data residency. Prompts on a custom-model account go to your provider, not to OpenAI. If that matters for your work, it matters in both directions.

Where we would tell you not to

  • If you are trying to get OpenAI’s models without paying for them, this does not do that and nothing here will help.
  • If your organization forbids third-party model endpoints, this is exactly the thing it forbids.
  • If you need the ChatGPT-hosted features, a custom-model account is the wrong tool — they are not available on it.

Related

Docs: Custom model providers Guide: run a custom model provider Using the Codex app without an OpenAI subscription Use an aggregator like OpenRouter Reasoning effort for custom models Is running multiple Codex accounts allowed?

Is this against OpenAI's terms of service?

We are not your lawyer and we do not speak for OpenAI. What we can state factually: Codex reads a documented custom model provider configuration, and the app ships non-ChatGPT authentication modes including API-key and Amazon Bedrock. A custom-model account uses those supported paths and consumes nothing from OpenAI. Read your own plan terms and decide for yourself.

Does CodexUse modify the Codex app?

No. It opens the app you installed with an isolated home directory and a configuration file, both of which the app supports by design. Nothing is patched, injected, or repackaged.

Will my ChatGPT account be affected?

No. Accounts are isolated and exclusive — an account is either an OpenAI account or a custom provider. Your ChatGPT logins, history, and quota are untouched by a custom-model account.