GPT-5.5 Will Be Disabled in Codex on October 14: What to Choose Instead

Additional instructions for working with Ailoot are in the documentation.

Which GPT-5.5 access in Codex is being retired?#

OpenAI gives the date and scope of the retirement on the official Codex models page:

Translation: On October 14, 2026, GPT-5.5 will be retired from ChatGPT, ChatGPT Work, and Codex on all plans, including consumer, Business, Enterprise, and Edu plans. This retirement does not apply to the OpenAI API.

In Codex, this affects sign-in with a ChatGPT account. If you authorize Codex with your own API key, access to models is determined by the OpenAI API organization and project associated with the key. This retirement does not affect them.

This is not the first such retirement. The same page says that gpt-5.4 and gpt-5.4-mini were already retired from Codex with ChatGPT sign-in on August 31, 2026, and GPT-5.3-Codex-Spark was retired on September 14, 2026. The GPT-5.5 announcement appeared in the official “What’s New” roundup for the week of September 14–18, 2026.

Changing the model in Codex affects several places at once: the settings file, saved settings, and launch commands.

Which Model Is the Recommended Replacement#

The Codex models page lists replacements by plan:

Translation: If you use Codex with ChatGPT sign-in, choose an available replacement before October 14. On Plus, Pro, Business, Enterprise, and Edu plans, that is GPT-6 Sol (gpt-6-sol) when available. On Free and Go, it is GPT-6 Luna (gpt-6-luna) in the desktop app when available.

At the time of writing (September 30, 2026), the newer GPT-6.1 Sol is also available. In its recommended models section, OpenAI writes:

Translation: For complex coding work and agentic workflows, use GPT-6.1 Sol if it is available to your account and client. Luna is suited to focused, repeatable tasks. GPT-6.1 Sol delivers results close to Astra on complex work at a lower cost than Astra.

The condition “when available” matters here. According to the same page, access to GPT-6.1 Sol is rolling out in stages: at launch, it is included in Plus, Pro, Business, Enterprise, and Edu plans in Codex in the desktop app and CLI, as well as in ChatGPT Work on the web and in the mobile app. Free and Go are not included at launch. In Enterprise and Edu, GPT-6.1 Sol is turned off by default until an administrator enables it.

OpenAI calls GPT-6 Sol and GPT-6 Luna the replacements for GPT-5.5. GPT-6.1 Sol is a newer model that OpenAI recommends for complex coding work where it is already available. Availability depends on the rollout stage, sign-in method, and client; in an organization, it also depends on whether an administrator has enabled the model.

How to Change the Model#

How you change the model depends on where you use Codex.

In an interactive CLI session (when working with Codex in the terminal), the /model command lets you change the model or reasoning effort during a conversation. You can also specify a model at launch:

Translation: In an interactive CLI session, use /model to change the model or reasoning effort. You can also choose a model when launching Codex with the --model flag or its shorter form, -m.

The documentation gives an example for GPT-6.1 Sol:

codex --model gpt-6.1-sol

If you have a Plus, Pro, Business, Enterprise, or Edu plan, GPT-6.1 Sol is not yet available, and GPT-6 Sol is already available to your account and client, specify the official replacement gpt-6-sol with the same flag:

codex --model gpt-6-sol

The same flag also works in non-interactive mode. Example from the documentation:

codex exec -m gpt-6.1-sol "Review the current changes"

With the official replacement for your plan, the command looks like this:

codex exec -m gpt-6-sol "Review the current changes"

For Free and Go, OpenAI specifies GPT-6 Luna as the replacement specifically in the ChatGPT desktop app, so on those plans, select it there using the model switcher.

In the ChatGPT desktop app (Work or Codex), ChatGPT Work on the web, and the IDE extension, the model and reasoning effort switcher is below the message input field.

Setting the default model through the settings file. The ChatGPT desktop app, Codex CLI, and IDE extension all read the same config.toml file. To set the model, replace the existing model value or add this setting if it is not already present. Example from the documentation:

model = "gpt-6.1-sol"

For the official replacement on Plus, Pro, Business, Enterprise, and Edu:

model = "gpt-6-sol"

For Free and Go, where the replacement is GPT-6 Luna:

model = "gpt-6-luna"

OpenAI says that gpt-6-luna can be selected in the model switcher or in the saved configuration when the model is available.

If no model is specified in config.toml, Codex uses one of the recommended models. This does not remove the need to check places where gpt-5.5 is explicitly selected: OpenAI asks users to replace it in the workspace’s default model settings, saved model settings, managed configurations, custom agents, scheduled tasks, launch scripts, and commands.

Additional Check for Enterprise and Edu#

If you are a workspace administrator or use Codex within an organization on an Enterprise or Edu plan, changing the model has an additional step. The official workspace model availability page warns:

Translation: At launch, GPT-6 Sol and GPT-6 Luna are turned off by default in Enterprise workspaces. An administrator must enable each model before members can select it.

GPT-6.1 Sol has its own rollout plan: in Enterprise and Edu, it is also turned off by default until an administrator enables it. Selecting a model in the local settings file does not grant access to it. Changing the workspace’s default model also does not grant access:

Translation: Changing the default model does not grant access to the model. Check that the replacement is available to the intended users in each client.

Model access (whether a user can select a model) and execution permissions (what an agent is allowed to do in the sandbox and which files and network destinations it can access) are set independently. The same page says:

Translation: Model access determines whether a model is available to the authenticated user in a supported client. Local permission profiles and managed requirements determine what an agent can do after a run starts, such as which files it can change and which network destinations it can reach.

If you specify gpt-6-sol or gpt-6.1-sol in config.toml but the administrator has not yet enabled that model for your workspace, you will not be able to select it. A permission profile does not grant model access. Conversely, access to a model does not loosen the sandbox, confirmation rules, or network restrictions.

What to Check Before October 14#

1. Determine how you sign in to Codex. If you sign in with a ChatGPT account (not a separate OpenAI API key) and use GPT-5.5, check your settings and choose an available replacement before the retirement.
2. Find explicit references to gpt-5.5 in config.toml, saved model settings, custom agents, scheduled tasks, launch scripts, and commands such as codex exec -m where the model is specified explicitly.
3. Replace it with a model available to you: GPT-6 Sol or GPT-6 Luna, depending on your plan, or GPT-6.1 Sol if it is already available to your account and client. If you work in an organization, first check with the administrator whether the needed model is enabled in the workspace.
4. Try the replacement on a small, familiar task. Seeing the model in the switcher alone is not enough for this check. OpenAI separately warns that reasoning levels do not match one-to-one across model generations and recommends starting with a lower level, then adjusting it based on the result.

What You Cannot Conclude from This#

The documentation does not promise that the new model will answer your specific tasks the same way as the old one. The models page points to official recommendations for choosing a model, where GPT-6.1 Sol is compared with Astra and Luna. Those guidelines do not replace testing on your project. This article presents OpenAI’s recommendations and does not include results from an independent model comparison.

The exact time when GPT-6 Sol, GPT-6 Luna, or GPT-6.1 Sol will become available for your specific plan and client is not guaranteed: according to OpenAI, availability depends on the rollout stage, sign-in method, and client. Selecting a model in the settings file also does not change the agent’s sandbox permissions; the official documentation treats these mechanisms separately.

If you are considering other tools, see the overview of Codex alternatives in 2026. There are separate articles about choosing a model: Which AI model to choose for a task and GPT-6 Astra or Claude Fable 5.1: Which Should You Choose. GitHub Copilot has its own retirement announcement: it is a different product with a different model list and its own date. See the article GitHub Copilot is retiring 6 models for details.

Before October 14, find where gpt-5.5 is explicitly selected and try the replacement on a small, familiar task.

For further work with models, you can go to Ailoot.