Blog / Troubleshooting

GPT-5.5 Retired in Codex: Find and Fix Old Model Names in 3 Steps

2026-10-19cdnping editors4 min read

GPT-5.5 was retired from Codex at noon PDT on October 14, 2026 (19:00 UTC, 3:00 a.m. on October 15 in Beijing). The change affects Codex sessions signed in with a ChatGPT account. Sessions that authenticate with an API key are not affected. If your config, scripts, or scheduled jobs hardcode gpt-5.5, the ChatGPT-login ones need to change to gpt-6.1-sol, or they may suddenly start failing.

This article is current as of October 2026 and is based on OpenAI's Codex changelog.

#What does the official changelog say?

The 2026-10-14 entry states:

• GPT-5.5 is retired from Codex as of that moment, for users signed in with ChatGPT.
• The OpenAI API and Codex sessions authenticated with an API key are not affected.
• The recommended replacement is GPT-6.1 Sol (gpt-6.1-sol), provided it is available on your plan and client.
• Places to check include: workspace default settings, saved model settings, managed configuration, custom agents, scheduled tasks, and scripts.

The same changelog also records that Codex CLI 0.159.1 (September 29) made GPT-6.1 Sol the default model in the built-in catalog and the Amazon Bedrock catalog, and 0.161.0 (October 7) noted the same default change again. So a freshly installed or upgraded CLI already uses the new default when you don't specify a model.

The changelog doesn't say what error an old model name returns, so this article doesn't guess at error text. Go by the actual output on your machine.

#Am I affected?

First, work out which login method you use.

GPT-6.1 Sol may not be available on every plan or client; check the official documentation for specifics. For usage limits, see our article How Codex usage limits work.

#How do I migrate in three steps?

#Step 1: Find every place that names the old model

Search your projects and home directory:

grep -rn "gpt-5\.5" . ~/.codex 2>/dev/null

Don't forget the places that live outside your repository: CI environment variables and secrets, crontab, systemd timers, team-shared managed configuration, and the default model setting for cloud workspaces.

#Step 2: Decide the fix based on your login method

Use a command like codex login status to confirm the current authentication method (check codex --help for the exact subcommand). Scripts that use ChatGPT login must be changed. Scripts that use an API key can be left alone for now, but if the same config runs in both environments, change both so behavior doesn't diverge.

#Step 3: Verify after the change

1. In your config, change model = "gpt-5.5" to model = "gpt-6.1-sol", or delete the line and let the CLI use its built-in default.
2. For codex exec scripts, check whether the old name is passed with the -m / --model flag, and replace it too.
3. Run a minimal task by hand, such as asking it to reply with a single sentence, and confirm it returns normally.
4. Trigger the scheduled job manually once. Don't wait until the next top of the hour to discover it is still failing.

For what changed in the CLI upgrade itself, see Codex retires its old default model on October 14, and CLI 0.163 adds new features.

#Still failing after the change: model problem or service problem?

First rule out leftover old names: run the grep above again, and watch for different capitalization and date-suffixed variants.

If the model name is correct and you still see timeouts or capacity errors, that is a different class of failure. Check cdnping's /en/down/codex and /en/status/openai. The latter is based on scheduled probes of the official API entry points from Cloudflare's edge; it only shows whether the entry point is reachable, not whether a specific model is available. Also see our log of October Codex errors and "Selected model is at capacity".

If you connect Codex through a custom endpoint, the model name depends on what that provider supports. See Claude Code, Codex, and OpenCode endpoint configuration for the syntax. A similar "old model name retired" problem is covered in what to change in your code now that deepseek-v4-flash is retired.

#How do I catch retirement notices earlier next time?

Model retirements usually show up in an official changelog first. cdnping's /en/changes monitors the official changelogs for OpenAI, Codex, and others and records additions and removals line by line. It has an RSS feed, so you can subscribe in your own reader. A more robust approach is to add a check to your repository: have CI grep for hardcoded model names and raise a warning, and keep model names in a single config file instead of scattering them across scripts.

#Sources

• OpenAI Codex changelog (GPT-5.5 retirement, 2026-10-14; CLI 0.159.1, 0.161.0)