Fix: gpt-5.3-codex Not Supported With ChatGPT

Lisa Ernst · 03.06.2026 · Developer Tools · 8 min read

If Codex suddenly returns "The 'gpt-5.3-codex' model is not supported when using Codex with a ChatGPT account", your installation is usually not broken. The client is trying to use a model that is not accepted through your current ChatGPT-account authentication path.

This guide shows the practical fix: select a supported Codex model, update the client, refresh authentication and avoid switching to API-key billing unless you deliberately want usage-based API access.

Source: Image source: ComputerProgrammer.jpg, Wikimedia Commons / Pixabay, CC0.

Quick fix

First, force Codex to use a model that is documented as a supported Codex choice for current workflows. OpenAI's Codex model documentation recommends starting most coding tasks with gpt-5.5, while gpt-5.4-mini is a faster fallback for lighter tasks.

codex -m gpt-5.5

If you need a lighter fallback, test this as well:

codex -m gpt-5.4-mini

Source: Screenshot recreated from the exact error text provided for this article.

The important part is not only the HTTP 400 status. The useful clue is the account-specific message saying that gpt-5.3-codex is not supported with a ChatGPT account.

What the error means

The message does not necessarily mean that gpt-5.3-codex never existed or that your Codex installation is corrupt. It means that the request path you are using right now does not allow that model with your ChatGPT-account session.

OpenAI documents different Codex authentication paths. ChatGPT sign-in uses your ChatGPT plan or workspace entitlement. API-key sign-in uses your OpenAI Platform account and is billed at API rates. These paths can have different model availability, billing and limits.

Source: Visual based on OpenAI Codex authentication documentation; OpenAI logo from Wikimedia Commons.

Do not treat ChatGPT-login Codex access and API-key access as the same thing. They are different authentication and billing paths.

Why it happens

This error is commonly caused by one of these situations:

• Your Codex client is still configured to use gpt-5.3-codex.
• The model is not enabled for your ChatGPT plan, workspace or current rollout state.
• The client falls back to a model that was previously available but is no longer accepted in your session.
• You are mixing up ChatGPT-plan access with OpenAI API-key access.
• Your local Codex session is stale after a model or entitlement change.

Step 1: Set a supported model in config.toml

If the explicit command works, persist the model in your Codex configuration file so Codex does not fall back to the unsupported model again.

# macOS / Linux
~/.codex/config.toml

# Windows
%USERPROFILE%\.codex\config.toml

model = "gpt-5.5"

Source: Screenshot created from the actual config command shown in this guide.

The safest permanent fix is to stop storing the unsupported model as your default and save a currently supported model instead.

Step 2: Update Codex

If your model list or routing behavior is stale, update Codex and restart your terminal or IDE. Depending on your installation method, use one of these commands.

# Windows
powershell -ExecutionPolicy ByPass -c "irm https://chatgpt.com/codex/install.ps1 | iex"

# npm
npm install -g @openai/codex

# Homebrew
brew install --cask codex

After the update, test again with an explicit model:

codex -m gpt-5.5

Step 3: Sign out and sign in again

If Codex still shows outdated models or keeps selecting the same failing model, refresh your ChatGPT authentication session. Sign out, restart the client and sign in again with the account that should have Codex access.

Important: do not upload, paste or screenshot your auth.json file in public tickets. Authentication files and tokens must be treated like secrets.

Step 4: Use API-key authentication only deliberately

API-key authentication can be useful for local CLI or IDE workflows, automation and CI/CD. But it is not the same as using included ChatGPT plan access. OpenAI's Codex authentication documentation states that API-key usage is billed through the OpenAI Platform account at standard API rates.

So the API-key route is not a free workaround. Use it only when you want platform billing and have checked usage limits.

Source: Visual based on OpenAI Codex model documentation; OpenAI logo from Wikimedia Commons.

For a normal troubleshooting article, the safest recommendation is to switch to a documented Codex model instead of repeatedly forcing the rejected gpt-5.3-codex name.

Public reports confirm the issue

The exact error has been reported publicly in the OpenAI Codex GitHub repository. One issue describes Codex CLI returning "The 'gpt-5.3-codex' model is not supported when using Codex with a ChatGPT account." after logging in with a ChatGPT Plus account. Similar reports mention gpt-5.4, gpt-5.3-codex and Codex app or VS Code extension workflows.

Source: Visual recreated from public GitHub issue #14735 text; GitHub Octicon from Wikimedia Commons.

The exact error text is a strong long-tail keyword because developers usually paste the full error into Google when Codex stops working.

Decision table

Final recommended setup

For most developers, the clean setup is simple:

• Use ChatGPT sign-in for normal Codex app, CLI or IDE usage.
• Set model = "gpt-5.5" in config.toml.
• Keep gpt-5.4-mini as a fast fallback.
• Update Codex before assuming your account is broken.
• Use API-key authentication only when you accept API billing.

Source: Image source: ComputerProgrammer.jpg, Wikimedia Commons / Pixabay, CC0.

This article uses real stock photography and source-based screenshots instead of generic AI-generated illustrations.

FAQ