# How do you move your Codex work off GPT-5.5?

Date: 2026-10-22 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/move-codex-work-off-gpt-5-5/

**Answer:** GPT-5.5 left Codex on October 14 for everyone signed in with a ChatGPT account, so any default, saved setting, custom agent, scheduled task or script that still names gpt-5.5 needs a new model, usually GPT-6.1 Sol (gpt-6.1-sol) on paid plans. Find every reference, swap it, rerun a few familiar tasks, and adjust reasoning effort from there.

On October 14, at noon Pacific time, GPT-5.5 was retired from Codex, OpenAI's coding agent, for users who sign in with a ChatGPT account. OpenAI announced the date a month ahead in the Codex changelog, and the same retirement applied to ChatGPT and ChatGPT Work. If you have opened Codex since then and seen a different model, or a task that suddenly fails, this is likely why.

Two groups are not affected. The OpenAI API still offers GPT-5.5, and Codex sessions that sign in with an **API key** (a secret code tied to a developer account and billed per use) keep access to it. Everyone on a ChatGPT plan needs to move.

## Which model to move to

OpenAI's recommended replacement on paid plans is **GPT-6.1 Sol**, released at DevDay on September 29 and described as close to its flagship GPT-6 Astra on coding and computer use at a fraction of Astra's price. Codex made it the default in its built-in model list the same day.

| Your situation | What to do |
|---|---|
| Plus, Pro, Business, Enterprise or Edu, signed in with ChatGPT | Switch to `gpt-6.1-sol` where it is available to you |
| Enterprise or Edu workspace | Ask your administrator to enable the new model before you change defaults |
| Signed in with an API key | Nothing is forced; GPT-5.5 still works, but plan your move anyway |

## The migration checklist

OpenAI's notice lists the places where `gpt-5.5` can hide. Work through them in order.

**1. Search for the old name.** From your project folder, and in your Codex settings folder, look for any text that still says `gpt-5.5`:

```bash
grep -rn "gpt-5.5" .
grep -rn "gpt-5.5" ~/.codex
```

Anything these commands find is a candidate for change: configuration files, scripts, notes for agents, or build tooling.

**2. Change your default model.** Codex reads a `config.toml` configuration file. One line sets the model used by default in the CLI and the IDE extension:

```toml
model = "gpt-6.1-sol"
```

**3. Update one-off commands.** Scripts and aliases that start Codex with a specific model need the new name:

```bash
codex --model gpt-6.1-sol
```

In an interactive session, the `/model` command lets you switch without restarting.

**4. Fix saved settings and workspace defaults.** In the ChatGPT desktop app or the IDE extension, open the model picker and pick the new model, then check any workspace default your team shares.

**5. Update custom agents and scheduled tasks.** These are easy to miss because they run without you. Open each one that names a model and change it. A scheduled job that quietly fails on a Monday night costs more time than the five minutes it takes to check now.

**6. Rerun three familiar tasks.** Pick tasks you have already done with GPT-5.5 and know the right answer to, such as fixing a known bug, writing a test, or reviewing a small change. Compare the result, the time taken, and how much of your usage allowance it consumed.

```prompt
In this repository, the date parser rejects dates written as 2026-10-05T09:00.
Write a failing test that shows the bug, fix it with the smallest change,
run the test suite, and summarize what you changed in five lines.
```

> **Try it:** Before changing anything else, run the two `grep` commands above on your main project and count how many places still name `gpt-5.5`. That number is the size of your migration.

## Why rerunning matters

A new model is not a drop-in copy of the old one. OpenAI reports that GPT-6.1 Sol beat GPT-6 Sol's best result on a long coding test while using a lower reasoning effort, and that it is more reliable at respecting explicit restrictions. Both are good news, but both can change how your prompts behave. A setting that suited GPT-5.5 may now be more than you need, and instructions that GPT-5.5 sometimes ignored may now be followed to the letter.

## Which settings should you check after the switch?

**Reasoning effort.** This controls how long the model thinks before acting. Don't assume the old level is right. Start with the default your client offers, try one step lower on the three tasks from the checklist, and keep the lower one if quality holds. You will save time and usage.

**Model access in shared workspaces.** In Enterprise workspaces, OpenAI leaves it to administrators to enable new models. If a teammate's Codex falls back to another model or refuses to start a task, check access before debugging their setup.

**Project-level files.** A project can carry its own configuration and instruction files, separate from your personal settings. If one project still behaves like the old setup, look for a model line in that project's own files.

**Automation and scripts.** Continuous-integration jobs, shell aliases and saved commands often name a model explicitly. Search them the same way you searched your project.

**Your sign-in method.** If you switch to an API key to keep using GPT-5.5 for a while, remember that usage is then billed to your developer account rather than counted against your ChatGPT plan. Make that choice deliberately, and set a date to finish the move.

## Sources

- Codex changelog: https://developers.openai.com/codex/changelog
- Introducing GPT-6.1 Sol: https://openai.com/index/introducing-gpt-6-1-sol/
- openai/codex README: https://github.com/openai/codex/blob/49342b156d55f0b7d7b7e63eae6702456966d998/README.md