GPT-5.5 retires on October 14, 2026: what ChatGPT and Codex users should switch to

OpenAI switches GPT-5.5 off on 14 October 2026. If you sign in to Codex with a ChatGPT account, switch to GPT-5.6 Sol (gpt-5.6-sol), which is the replacement OpenAI names on both its changelog and its models page; if you mostly chat, GPT-5.6 Terra is the model OpenAI describes as competitive with GPT-5.5 at lower cost. The OpenAI API is not affected, and the things most likely to break quietly are the settings nobody looks at: workspace defaults, custom agents and scheduled tasks that still name gpt-5.5.

#What is retiring, and when?

OpenAI published the deadline twice in the same week. The What's new entry for the week of 14 to 18 September 2026 states:

GPT-5.5 will retire from ChatGPT, ChatGPT Work, and Codex on October 14, 2026, across all plans.

The ChatGPT and Codex changelog entry dated 14 September 2026 spells out which plans that covers:

On October 14, 2026, GPT-5.5 will retire from ChatGPT, ChatGPT Work, and Codex on all plans, including consumer, Business, Enterprise, and Edu plans.

Both entries say the retirement does not affect the OpenAI API. That is the single most useful sentence in the announcement, because it means a production integration built on API keys keeps working while a colleague's scheduled task in the ChatGPT app does not.

#What should you switch to?

OpenAI's models page names the replacement directly for the Codex case:

If you use Codex with ChatGPT sign-in, switch to GPT-5.6 Sol (gpt-5.6-sol) before October 14.

The changelog says the same thing in the other direction: switch from gpt-5.5 to gpt-5.6-sol. These are the models the page lists above GPT-5.5:

Model	Identifier	What OpenAI's models page says it is for
GPT-6 Astra	gpt-6-astra	Most capable model for complex work across code, apps and research
GPT-5.6 Sol	gpt-5.6-sol	Most capable GPT-5.6 model for complex coding, computer use, research and cybersecurity
GPT-5.6 Terra	gpt-5.6-terra	Balanced GPT-5.6 model for everyday work, competitive with GPT-5.5 at lower cost
GPT-5.6 Luna	gpt-5.6-luna	Fast and affordable GPT-5.6 model with strong capability at lowest cost

Sol is the model OpenAI tells Codex users to move to, so that is the safe default for coding work. Terra is the one the page positions against GPT-5.5 itself, at lower cost, which makes it the closer match for general chat and document work. If the model picker matters to your budget, our cost optimization guide covers the general shape of the tradeoff, and the ChatGPT hub carries the current plan prices.

#What do you need to change before 14 October?

The retirement notice asks you to update saved model settings, workspace defaults, custom agents, scheduled tasks and scripts. In practice that is five different places, and only the first is obvious.

1. Your saved model choice in the ChatGPT apps. Wherever you currently select GPT-5.5 by hand, select one of the models below instead. This one fixes itself the moment you notice, which is why it is the least of the problem.
2. The Codex CLI. The developer commands reference gives /model to change the model mid-session, and the --model, -m flag to override the configured model for a single invocation.
3. Custom agents and subagents. OpenAI's subagents documentation describes agent files under ~/.codex/agents/ for personal agents and .codex/agents/ for project-scoped ones, each a standalone TOML file, with model and model_reasoning_effort keys. Defaults live in config.toml under [agents], as agents.default_subagent_model and agents.default_subagent_reasoning_effort.
4. Scheduled tasks. The scheduled tasks page says you can leave the model and reasoning effort on their defaults or choose them explicitly, and it asks you to review the tasks that use GPT-5.5 and choose an available replacement before the date. A task that runs weekly is the one that will fail while you are not looking.
5. Workspace defaults, if you administer a workspace. The workspace model availability page says owners and admins can configure separate starting defaults for Chat and for Work and Codex, and asks admins to review those defaults for all three surfaces before 14 October. The same page is careful that these are defaults and not grants: starting defaults do not grant access to unavailable models or override access requirements, so setting a default is not the same as giving your members a model.

Anything that names the string gpt-5.5 outside those five places, in a shell script, a CI job or a saved prompt, needs the same edit. A plain search across your own repositories for the literal model id is the cheapest version of this audit.

#Does this affect the OpenAI API?

No. Both primary entries carve the API out explicitly. If your agent talks to OpenAI with an API key rather than a ChatGPT sign-in, this deadline is not yours, though the models available to a key follow that key's own API model list rather than the ChatGPT plan table. The distinction matters for the Codex hub case in particular, because Codex can be driven either way, and only the ChatGPT sign-in path is covered by the retirement.

#Has OpenAI done this before?

Yes, six weeks earlier. The models page records that GPT-5.4 retired from Codex with ChatGPT sign-in on 31 August 2026, and the scheduled tasks page carries the matching instruction for task owners: if a scheduled task uses gpt-5.4 or gpt-5.4-mini with ChatGPT sign-in, update it before those models retire on 31 August 2026. API access was unaffected there too.

Two retirements in seven weeks is the useful pattern here, and it is worth planning around rather than reacting to. Pinning a model id in a script buys reproducibility now and an outage later; naming a model family and re-checking it each quarter costs a calendar reminder. If you are weighing a hosted assistant against something you run yourself, that maintenance difference is one of the honest inputs, and we cover the rest of it in OpenClaw vs ChatGPT.

#What we could not confirm

We did not find a page stating what happens on 15 October to a scheduled task, custom agent or workspace default that still names gpt-5.5. The pages we read (What's new, the changelog, models, workspace model availability and scheduled tasks) all ask you to choose a replacement before the date, and none of them describes the fallback behaviour if you do not. Treat "it will silently fall back to something sensible" as an assumption nobody has published, and change the setting.