---
title: "GPT-5.5 Leaves Codex on October 14. Here's What to Do Before Then"
author: "Ecem Karaman"
source: "https://aiwithecem.com/guides/gpt-5-5-retirement"
published: 2026-09-19
tools: ["Codex","ChatGPT"]
topics: ["Setup","Workflows"]
---

# GPT-5.5 Leaves Codex on October 14. Here's What to Do Before Then

On October 14, 2026, OpenAI retires GPT-5.5 from ChatGPT, ChatGPT Work and Codex. If anything in your setup depends on it, here's what to expect, how to find those dependencies, and your options for keeping things running: migrate to a newer model, or keep GPT-5.5 through the API.

**Details checked against [OpenAI's Codex models page](https://learn.chatgpt.com/docs/models) and [workspace model availability docs](https://learn.chatgpt.com/docs/enterprise/workspace-model-availability) on September 19, 2026.**

> **In this guide**
>
> 1. [What's changing](#whats-changing)
> 2. [Sol or Terra?](#sol-or-terra)
> 3. [Where gpt-5.5 hides](#where-gpt-55-hides)
> 4. [Test before the deadline](#test-before-the-deadline)
> 5. [Pros and cons](#pros-and-cons)
> 6. [The bigger lesson](#the-bigger-lesson)
> 7. [Before October 14](#before-october-14)

## What's changing

| Where | What happens on October 14 |
| ----- | -------------------------- |
| ChatGPT, ChatGPT Work, and Codex signed in with ChatGPT | GPT-5.5 is gone, on every plan: consumer, Business, Enterprise and Edu |
| The OpenAI API | Not affected. `gpt-5.5` stays available |
| OpenAI's official replacement | **GPT-5.6 Sol** (`gpt-5.6-sol`) |

## Sol or Terra?

OpenAI's migration guidance says to switch to Sol. Its own model guide also says **Terra** is "a natural starting point for work you previously gave GPT-5.5," with performance competitive with GPT-5.5 at a lower cost.

| Pick | When |
| ---- | ---- |
| **GPT-5.6 Sol** | You used GPT-5.5 for hard, open-ended work: complex code changes, deep research, polished documents |
| **GPT-5.6 Terra** | You used it as your everyday default and want similar results for less usage |
| **GPT-6 Astra** | The hardest multi-step work across code, apps and research, where your plan includes it |

**Also note:** GPT-5.5's reasoning effort levels don't map exactly to GPT-5.6. OpenAI suggests trying a familiar task at a **lower** setting first, then adjusting.

## Where gpt-5.5 hides

Changing the model in the picker isn't enough if you set it anywhere else. OpenAI lists these places:

- [ ] Workspace defaults
- [ ] Saved model settings
- [ ] Your `config.toml` (`model = "gpt-5.5"`)
- [ ] Custom agents
- [ ] Scheduled tasks
- [ ] Scripts and commands that select a model, like `codex exec --model gpt-5.5`

To find hardcoded references in your config and current project:

```bash
grep -rn "gpt-5\.5" ~/.codex/config.toml . 2>/dev/null
```

**Admins:** changing a workspace default doesn't grant access to the new model. Confirm the replacement is available to each affected user and client.

## Test before the deadline

Don't find out on October 15 that a scheduled task now behaves differently. Pick three tasks you run often, then:

```text
Here's a task I used to run on GPT-5.5: [task + example output]. Do it again. Then tell me what you did differently from that example and what I should change in my prompt to get the result I expect.
```

Run it on Sol and Terra at a lower effort than you used before. Keep whichever meets your bar at the lower cost.

## Pros and cons

| Pros | Cons |
| ---- | ---- |
| Newer models, and Terra may match your GPT-5.5 work at a lower cost | Hardcoded settings and scheduled tasks can break or shift quietly |
| A simpler lineup: Astra, Sol, Terra and Luna | Prompts tuned for GPT-5.5 may behave differently |
| The API keeps GPT-5.5 if you truly depend on it | Staying on it through the API means paying per token instead of using your plan |

## The bigger lesson

GPT-5.5 launched in April and leaves ChatGPT and Codex in October. GPT-5.4 left Codex on August 31, and GPT-5.2 and GPT-5.3-Codex were deprecated before that. Model retirements are now a regular event, not a rare one.

- **Set the model in one place.** Use a single `config.toml` line or one default, not a model name scattered across scripts and agents.
- **Test with tasks, not vibes.** Keep a few saved tasks with known-good outputs, so every switch takes minutes.
- **Keep your system bigger than any model.** Your context, skills and workflows should outlast the model underneath. [Rebuild your AI system](/guides/weekend-ai-system) so any tool or model can plug in.

## Before October 14

- [ ] Search for `gpt-5.5` and replace it everywhere
- [ ] Test three real tasks on Sol and Terra
- [ ] Update scheduled tasks and custom agents
- [ ] Workspace admins: confirm access to the replacement
- [ ] Only if you truly need GPT-5.5: plan for API access and its costs

**Go deeper:** [OpenAI's Codex models page](https://learn.chatgpt.com/docs/models) · [Claude Code vs Codex](/guides/claude-code-vs-codex) · [Stop Building Your AI System Inside One Tool](/guides/tool-agnostic-ai-system)