GPT-5.5 Retirement on Oct 14: What Changes, What to Use

OpenAI is pulling GPT-5.5 out of its apps eight days from now. For most people, that means one less entry in a menu. For anyone who wired the name gpt-5.5 into a Codex config, a scheduled task or a team default, it’s a deadline. Here’s what changes, who it hits, and what I’d switch to.

All dates, model names and prices below come from OpenAI’s own docs, checked on October 6, 2026. We haven’t run our own side-by-side tests on the replacement models yet, so any quality claims here are OpenAI’s, not ours.

What’s changing, and exactly when

OpenAI posted the notice in its ChatGPT & Codex changelog in an entry dated September 14, 2026. The wording is short:

• Date: October 14, 2026. OpenAI hasn’t given a time of day, so treat October 13 as your real deadline.
• Where: ChatGPT, ChatGPT Work and Codex.
• Plans: all of them, “including consumer, Business, Enterprise, and Edu plans.”
• Not affected: the OpenAI API. The Codex models page adds that Codex run with your own API key isn’t affected either.

So this is a product retirement, not an API shutdown. The gpt-5.5 model ID keeps working for developers. It doesn’t appear on OpenAI’s API deprecations page, and no API end date has been announced.

It fits a pattern: GPT-5.4 and GPT-5.4 mini left Codex with ChatGPT sign-in on August 31, and GPT-5.3-Codex-Spark went on September 14. Older models don’t last long in OpenAI’s apps.

Who’s affected

The people most likely to get caught out are Codex users with things that run unattended. A scheduled task or a custom agent that names gpt-5.5 won’t warn you ahead of time. It’ll just fail on the day.

What happens to your old GPT-5.5 chats?

OpenAI hasn’t said exactly which model old GPT-5.5 conversations will move to. Its general help article, What to expect when models change, says prior conversations “can carry over to newer models,” and that tone or style may change. When GPT-5 left ChatGPT in February, older chats moved to GPT-5.5 equivalents. A similar move to the GPT-5.6 family seems likely, but OpenAI hasn’t confirmed it for this retirement.

In practice, your history isn’t deleted. Continued threads may just read differently.

What to switch to

There’s one catch that confuses people: OpenAI’s GPT-6 models (GPT-6 Sol, GPT-6.1 Sol and GPT-6 Luna) are available in Work and Codex, not in Chat. In the Chat experience, the replacement is the GPT-5.6 family.

OpenAI’s guidance for Codex has also shifted. Its September 15 social post told Codex users to switch to GPT-5.6 Sol or GPT-6 Astra. The current models page says something more specific: on Plus, Pro, Business, Enterprise and Edu, choose GPT-6 Sol (gpt-6-sol) “when available”. On Free and Go, choose GPT-6 Luna (gpt-6-luna) in the desktop app. For complex coding, the same page recommends GPT-6.1 Sol, which launched in Codex and Work on September 29.

API prices below are OpenAI’s standard rates per million tokens (input / cached input / output) for prompts under 272K tokens, from the API pricing page and model pages.

My pick: in Codex, use GPT-6.1 Sol if it shows up in your picker, and GPT-6 Sol if it doesn’t. Use Luna for repetitive, clearly defined jobs. Save Astra for the work where getting it wrong costs real money. Going by OpenAI’s own Codex credit rate card, GPT-6 Sol and GPT-6.1 Sol cost 50 credits per million input tokens and 250 per million output, compared with 125 and 750 for GPT-5.5. On credit-based billing, the same job should cost fewer credits. OpenAI notes that credit rates alone don’t determine how fast you use your included limits, so check your usage dashboard after the switch.

What to do before October 14

If you just use ChatGPT

1. Open the model menu or slider. If you’re still choosing GPT-5.5 by hand, move to the default (GPT-5.6 Sol on paid plans) and run a few of your usual prompts.
2. If you rely on a particular tone, put it in custom instructions so you aren’t depending on one model’s personality.

If you use Codex with a ChatGPT sign-in

1. Check how you sign in. ChatGPT sign-in is affected. API-key mode isn’t.
2. Search for the string. Look for gpt-5.5 in your config.toml, project configs, shell scripts and CI jobs. The desktop app, CLI and IDE extension all read the same config.toml.
3. Change the default. Set model = "gpt-6.1-sol" (or "gpt-6-sol") in config.toml. In the CLI you can also use /model, or codex --model gpt-6.1-sol and codex exec -m gpt-6.1-sol for one-off runs.
4. Check the things nobody opens. OpenAI’s checklist names workspace defaults, saved model settings, managed configurations, custom agents, scheduled tasks and scripts. Scheduled tasks and custom agents are the ones that fail without anyone noticing.
5. Re-run one real task. OpenAI says reasoning effort levels “don’t map exactly between model generations,” so compare a familiar job at a lower setting and adjust from there.

If you’re a workspace admin

1. Review the defaults for ChatGPT, Work and Codex separately, following OpenAI’s workspace model availability guide.
2. Enable the replacement first. In Enterprise and Edu, new models like GPT-6.1 Sol are off until an administrator turns them on. Changing a default doesn’t give anyone access.
3. Then move managed configurations from gpt-5.5 to gpt-6-sol, and tell your team the date.

If you build on the OpenAI API

Nothing breaks on October 14. Still, a few notes for your migration planning:

• Pin a snapshot if you need stable behavior. The current snapshot is gpt-5.5-2026-04-23 (model page).
• Watch the deprecations page. OpenAI’s policy is at least six months’ notice before retiring a generally available API model, so you’ll get a warning before anything changes.
• Do the cost math anyway. At list prices, 10M input and 2M output tokens a month comes to $110 on GPT-5.5 and $40 on GPT-6 Sol or GPT-6.1 Sol. That’s a big enough gap to justify running your evals now.
• Test before you swap. GPT-6 Sol supports the same 1,050,000-token context window. In Chat Completions, though, it supports function calling only with reasoning_effort set to none, so agent code should use the Responses API.
• Long prompts cost more. On GPT-5.5, GPT-5.6 Sol and GPT-6 Sol, prompts over 272K input tokens are billed at 2x the input rate and 1.5x the output rate.

Verdict

This isn’t a crisis, but it’s a real deadline for one group: Codex users who signed in with ChatGPT and haven’t touched their configs in a while. Spend ten minutes searching for gpt-5.5 this week. Chat-only users can mostly ignore it, because OpenAI already moved the defaults in August. API developers aren’t forced to do anything, but the price gap between GPT-5.5 and GPT-6 Sol is wide enough that I’d start testing now rather than wait for a deprecation email.

FAQ

Is GPT-5.5 being shut down completely?

No. It’s leaving ChatGPT, ChatGPT Work and Codex with ChatGPT sign-in. It stays available in the OpenAI API, and in Codex when you authenticate with an API key.

What time on October 14 does GPT-5.5 retire?

OpenAI hasn’t published a time or time zone. Finish any migration by October 13 so you don’t depend on when it actually happens.

Can I keep using GPT-5.5 in Codex after the retirement?

Only through an API key, and that usage is billed at API rates ($5 per million input tokens and $30 per million output at standard pricing) rather than coming out of your ChatGPT plan.

Why can’t I find GPT-6 Sol in regular ChatGPT?

Because OpenAI only offers GPT-6 Sol, GPT-6.1 Sol and GPT-6 Luna in Work and Codex. In Chat, paid plans get GPT-5.6 Sol, and eligible plans can pick GPT-6 Pro (Astra) under the Pro option.

Do Free and Go users lose anything?

Not much. Their ChatGPT default has been GPT-5.6 Luna since August. In the Codex desktop app, OpenAI points them to GPT-6 Luna.

Will the GPT-5.5 API model be retired soon?

There’s no announced date. OpenAI’s policy gives at least six months’ notice for generally available models, and GPT-5.5 isn’t on the deprecations list as of October 6, 2026.

Sources checked on October 6, 2026. We’ll update this post if OpenAI changes the date or the replacement guidance.