Reference

Check your ChatGPT, Codex, and API access

ChatGPT + Codex + OpenAI API

Identify the account and billing path actually used by a developer task, including Codex hosted inside an IDE.

Applies to
ChatGPT + Codex + OpenAI API
Last verified
Reviewed by
Timothy Fehr

Check the active product and authentication path before interpreting a quota message or cost estimate. An IDE can host Codex while the underlying session uses your own OpenAI account. The window's branding does not settle billing.

Identify the four relevant facts

Record the product, host, authentication method, and account or organization. Keep credentials out of that record.

SituationWhat to inspect
ChatGPT taskSigned-in account, selected workspace, available tools, and usage notice
Local Codex or IDE extensionAgent identity, local authentication status, configured model, and effective policy
Codex cloud taskChatGPT sign-in and the selected cloud environment
Direct API programAPI project/organization, configured credentials, model ID, and usage dashboard
IDE-provided AI serviceWhether the host supplies its own entitlement or connects your account

Codex's supported local authentication paths include ChatGPT sign-in and API credentials. Cloud use requires ChatGPT sign-in. Availability and policy depend on the actual access path and account.

Work through an IDE example

Suppose an IntelliJ panel says “Codex” and a task runs against your checkout. Inspect its agent settings and account status. If it reports ChatGPT authentication, record that separately from “host: IntelliJ”.

Then check the selected model and reasoning control in that host. A prompt requesting maximum reasoning does not prove that a configuration setting changed. If the setting is unavailable, record the limitation.

Do not inspect or share token contents to answer this question. A status screen or redacted configuration identity is the evidence you need.

Interpret usage in the right unit

A ChatGPT subscription entitlement is not a balance of direct API credits. API token rates do not predict a fixed number of Codex tasks on a subscription. Agent tasks differ in context, tools, reasoning, execution duration, and account limits.

For an automated pipeline, record each adapter's access method separately. An API stage and a locally authenticated Codex stage can have different accounting even when both use OpenAI models.

The model catalog gives selected API specifications and rates. Use the account's own current usage interface for the active product.

What goes wrong

A developer adds an API key to solve a local login problem and unintentionally changes the billing path. A team assumes its IDE licence covers every separately installed agent. A pipeline reports an API estimate as if it were subscription quota usage. Check the actual configuration after any authentication change.

How to check

Write a short record such as:

Product: Codex
Host: my IDE
Authentication: ChatGPT sign-in, confirmed in account status
Model / effort: values displayed by this host
Execution: local checkout
Usage evidence: the corresponding account usage view

Run a small synthetic task only after the access path is the one you intend. Keep the output and checks, then compare the usage notice in that account. Never infer a different payer from the model's answer about itself.

Sources

  1. OpenAI: authentication Tier 1 2026-09-08
  2. OpenAI: non-interactive Codex Tier 1 2026-09-08
  3. OpenAI: API pricing Tier 1 2026-09-08