GPT-5.5 Retired Today (Oct 14): Fix Codex & ChatGPT Errors

on

GPT-5.5 retired — fix Codex and ChatGPT errors banner

Sources

OpenAI pulled the plug today. As of October 14, 2026, GPT-5.5 is retired from ChatGPT, ChatGPT Work, and Codex across all plans — exactly as OpenAI's official ChatGPT account announced on September 15: "On October 14, it's time to say farewell to GPT-5.5 in ChatGPT, ChatGPT Work, and Codex across all plans." If your agents, scripts, or scheduled tasks still name gpt-5.5, they are failing right now. This guide fixes every common breakage, including the cryptic "not supported" error many Codex users hit while migrating.

Sources

What actually retired — and what didn't

The retirement covers GPT-5.5 and GPT-5.5 Thinking in three product surfaces: ChatGPT, ChatGPT Work, and Codex. It applies across all plans, including Free. The official replacement guidance for Codex users is to switch to GPT-5.6 Sol or GPT-6 Astra.

Sources

Crucially, the OpenAI API is not affected. Anything authenticated with an API key keeps working as usual — the cutoff is a product-surface change only. Do not assume an API deadline from this announcement; check the current API model documentation if you call models by key. Also note that everyday ChatGPT conversations were auto-migrated to OpenAI's new default, so most casual users saw nothing change — developers with hardcoded model names are the ones feeling the breakage.

Sources

Fix 1: Migrate everything hardcoded to gpt-5.5

Any request that names gpt-5.5 directly inside ChatGPT, Work, or Codex now fails instead of returning a response — there is no graceful degradation. Walk through this checklist:

Sources

  1. Codex CLI flags: search your scripts for --model gpt-5.5 and change it to --model gpt-5.6-sol or gpt-6-astra.
  2. config.toml / workspace defaults: update the saved model in your Codex configuration and any managed workspace defaults.
  3. Custom agents and scheduled tasks: agents with pinned model settings in Codex or Work must be re-pointed; scheduled tasks configured for GPT-5.5 will error on their next run.
  4. Plugins, connectors, and scripts: anything that stored the model string in environment variables or config files needs the new value.

After migrating, run a quick regression pass before declaring victory — OpenAI does not guarantee identical tool behavior, latency, or output format on the replacements. Notably, GPT-5.5 had a real following among power users for being fast and predictable on routine work, while GPT-5.6 Sol can over-reason on simple tasks, so budget extra testing time for prompt-sensitive workflows. (If you're weighing frontier alternatives, our frontier-model comparison covers what else launched this week.)

Sources

Fix 2: "The 'gpt-6.1-sol' model is not supported when using Codex with a ChatGPT account"

This is the error message currently confusing the most people, and its cause is not obvious: the ChatGPT backend only serves GPT-6.1 Sol to Codex clients version 0.159.0 and newer. If you are on an older CLI, requests to gpt-6.1-sol return HTTP 400 with the "not supported" message even on eligible plans — it looks like an account problem, but it is a version problem.

Sources

Developers confirmed this by probing the backend live: a client identifying as 0.155.0 never even sees gpt-6.1-sol in the model list, while 0.162.1 streams replies normally. The fix:

Sources

  1. Update your Codex CLI: npm i -g @openai/codex (or reinstall the Codex desktop/VS Code extension to pull its bundled CLI).
  2. Restart the CLI or app server so the new client version registers — old sessions keep the old identity.
  3. Retry. If the error persists on the latest version, the issue is plan eligibility instead: GPT-6.1 Sol requires a supported plan, and on unsupported tiers you should switch to a model your account can actually use rather than changing billing or configuration repeatedly.

Do not "fix" this by downgrading your client to keep an old model — retired models are not coming back, and stale clients accumulate exactly these failures.

Sources

Fix 3: Today's Codex and ChatGPT Work outage (Oct 14)

If everything was working yesterday and broke today, you may not have a configuration problem at all. OpenAI logged degraded performance in Codex and ChatGPT Work on October 14, with users seeing elevated error rates for roughly 2.8 hours (about 07:05–09:53 UTC) before a mitigation was applied and services recovered. October 13 saw a separate 8.5-hour degraded-performance incident on GPT-6.1 Sol across Codex and Work mode.

Sources

Rule of thumb during these windows: check status.openai.com before you touch your own setup. If the status page shows an active incident on Codex or ChatGPT, retrying in a loop or rewriting your config only burns quota and patience. Wait for the all-clear, then retest once.

Sources

Fix 4: The Thinking slider changed — "think harder" no longer switches models

Since September 14, ChatGPT Plus and Pro no longer auto-switch from Instant to Thinking when you ask for deeper reasoning, and the Higher intelligence setting is gone from ChatGPT on the web. In practice: typing "think harder" or "think deeply" will not change the thinking level anymore.

Sources

Set the thinking level yourself in the model picker: the slider runs from Instant through Medium, High, and Extra High (all powered by GPT-5.6 Sol) up to a Pro level. Plus gets Medium and High only; Pro, Business, and Enterprise get the full ladder. Free and Go users get a Think button instead, which runs the smaller GPT-5.6 Luna. If responses suddenly feel shallower than you remember, this setting — not a bug — is probably why.

Sources

Fix 5: Codex extension blank after a VS Code update (WSL users)

A regression reported October 13 (tracked against microsoft/vscode#341408) breaks the Codex extension's webview when VS Code runs through WSL with a distribution name containing uppercase letters, like Ubuntu. Symptoms: the Codex panel never loads, logs show HTTP 401 resource failures followed by renderer_ready_timeout — even though the Codex backend itself initializes fine. It works normally in VS Code on Windows.

Sources

Temporary workaround from the OpenAI developer community: force a lowercase remote authority when launching:

Sources

code --folder-uri "vscode-remote://wsl+ubuntu$(pwd)"

Run that from your Ubuntu terminal inside the project directory. This is a stopgap until Microsoft or OpenAI ships the real fix — keep your VS Code updated and retest with a plain code . after each update.

Sources

Frequently asked questions

Is GPT-5.5 gone from ChatGPT completely?

Yes, as of October 14, 2026 — GPT-5.5 and GPT-5.5 Thinking are retired from ChatGPT, ChatGPT Work, and Codex across all plans. Regular chats were auto-migrated to OpenAI's new default model. The OpenAI API was not part of this retirement.

Sources

What should I use instead of GPT-5.5 in Codex?

OpenAI's official guidance is to switch to GPT-5.6 Sol or GPT-6 Astra. Update --model flags, config files, saved workspace settings, custom agents, and scheduled tasks that reference gpt-5.5, then test each workflow before relying on it.

Sources

Why does Codex say "gpt-6.1-sol is not supported with a ChatGPT account"?

That error almost always means your Codex client is older than version 0.159.0 — the backend only serves GPT-6.1 Sol to newer clients. Update the CLI (npm i -g @openai/codex), restart it, and retry. If you are on the latest version, check that your plan supports the model.

Sources

Is Codex down right now?

Codex and ChatGPT Work had degraded performance on October 14, 2026 (elevated errors for about 2.8 hours) plus a separate 8.5-hour incident on October 13. Check status.openai.com for the live status before changing your own configuration.

Sources

Can I keep using GPT-5.5 as a legacy option in Codex?

No. OpenAI declined community requests to keep it as a legacy option, so there is no supported way to keep GPT-5.5 in ChatGPT, Work, or Codex. Migrate to GPT-5.6 Sol or GPT-6 Astra — the only path the company supports.

Sources


Comments

Post a Comment