GPT-5.4 and GPT-5.4 mini retire from Codex on 31 August 2026 for ChatGPT users

If you use OpenAI Codex through your ChatGPT account, there is a deadline you need to know about: 31 August 2026. After that date, GPT-5.4 and GPT-5.4 mini will no longer be available in Codex for users who sign in via ChatGPT. The replacements are GPT-5.6 Terra and GPT-5.6 Luna respectively.

This is a targeted change. If you access Codex through the API using an API key, nothing changes for you right now.

What exactly is changing

OpenAI published this as a formal changelog entry on 31 July 2026 at developers.openai.com/codex/changelog. The retirement applies only to ChatGPT-authenticated Codex sessions across Free, Go, Plus, Pro, Business, Enterprise, and Edu plans.

The model slug mappings are straightforward:

gpt-5.4 should be replaced with gpt-5.6-terra
gpt-5.4-mini should be replaced with gpt-5.6-luna

GPT-5.4 was introduced in Codex on 5 March 2026, giving it roughly a six-month run before this retirement. The GPT-5.6 family, which includes Sol, Terra, and Luna, became generally available from 9 July 2026.

What you actually need to update

Before 31 August 2026, go through anything in your Codex setup that references either model and update it:

Workspace default model settings
Saved model configurations
Custom agents you have built
Scheduled or recurring tasks
Any managed configuration files, including requirements.toml

OpenAI has already migrated its own internal uses and bundled GPT-5.4 selections to the GPT-5.6 equivalents, and the OpenAI Docs skill has been updated to reflect current model resolution. Auto-review in both the ChatGPT app and Codex CLI has also moved from GPT-5.4 to GPT-5.6 Luna.

If you are a workspace admin or owner, you can manage model availability and defaults from Workspace Settings → Models. The hierarchy to be aware of: a managed requirements.toml overrides workspace starting defaults, which in turn override a member’s local choice.

The GPT-5.6 family, briefly explained

The three models in the GPT-5.6 family each occupy a distinct position:

Sol is the flagship. It offers the highest reasoning ceiling and is designed for complex tasks, large codebases, and demanding long-running agentic work.

Terra is the balanced default, comparable in positioning to what GPT-5.4 was doing for most users. It is a strong all-round choice for everyday interactive and agentic coding.

Luna is the lightweight, cost-efficient option, built for smaller and faster tasks where you want results quickly and cheaply. This is the direct replacement for GPT-5.4 mini.

On 30 July 2026, OpenAI cut the price of GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20%. The practical upshot: Auto-review running on Luna is expected to cost around ten times less than it did on GPT-5.4. The migration is not just a feature upgrade; it is a meaningful cost improvement for anyone using Codex at volume.

In terms of navigation within Codex, the default Power setting uses GPT-5.6 Sol with medium reasoning. From there, “Smarter” increases reasoning depth and “Faster” reduces cost. You can also open Advanced settings to select gpt-5.6-luna or any specific model directly.

API users: you are not affected (for now)

To be precise: GPT-5.4 and GPT-5.4 mini remain available on the OpenAI API and in Codex sessions authenticated with an API key. The retirement announced here is scoped specifically to ChatGPT-authenticated Codex access.

That said, it is worth noting that OpenAI has also deprecated Support for the Chat Completions API within Codex, flagging that it will be removed in a future release. If your integration relies on that, keep it on your radar even if the August deadline does not affect you directly.

A couple of other changes worth noting

Two additional updates shipped alongside this retirement notice:

Fast mode replaces the previous Priority Processing offering in the API. For GPT-5.6 Sol, Fast mode delivers up to 2.5 times faster processing than standard, at twice the price. Existing requests tagged priority will automatically route to Fast mode, so there is no immediate action required on that front.

Workspace admin controls have been expanded. Admins can now configure the starting model, reasoning level, speed, Fast Mode availability, and new-chat behaviour from a single settings panel. Fast Mode is enabled by default.

The short version

If you use Codex through ChatGPT, check your saved configurations and custom agents before 31 August 2026. Swap gpt-5.4 for gpt-5.6-terra and gpt-5.4-mini for gpt-5.6-luna. You are moving to newer, cheaper models, and in most cases the transition should be straightforward. If you access Codex via the API with an API key, this particular deadline does not apply to you.

The full changelog is at developers.openai.com/codex/changelog, and the model details are at developers.openai.com/codex/models.