GPT-5.5 retires from Codex on October 14: check these model settings

OpenAI says GPT-5.5 will retire from ChatGPT, ChatGPT Work, and Codex on 14 October 2026. That date matters if a saved Codex setting, custom agent, or scheduled task still selects gpt-5.5. An interactive model picker may be easy to change; a forgotten automation may be harder to notice until its next run. The retirement does not apply to the OpenAI API.

Find the places that choose a model

OpenAI's migration guidance names workspace defaults, saved model settings, managed configurations, custom agents, scheduled tasks, and scripts. Inventory those places before changing a team default. Search the configuration files and task definitions you control for the exact model ID gpt-5.5, then record the owner and purpose of each match. A prose mention in a document is different from an executable setting; change the settings that actually select the model.

For teams with an administrator, check workspace model availability and managed defaults too. A local replacement setting cannot make a model available if the workspace has not enabled it. Keep an existing task and its acceptance checks as a comparison case, especially when the task uses tools or a long conversation.

Choose the replacement that your plan can use

OpenAI’s retirement guidance points Work and Codex users with ChatGPT sign-in to GPT-6 Sol (gpt-6-sol) once an administrator enables it. GPT-6 Luna (gpt-6-luna) is also rolling out in the desktop app for Free and Go users. Sol is positioned for complex coding and agentic work; Luna for focused, repeatable tasks. Availability varies by client, sign-in method, plan, workspace settings, and rollout. Check the model picker or administrator controls before setting a new default.

Do not turn this into an untested search-and-replace across every workflow. Run a representative saved task after changing its model. Verify the final diff, tests, external sources it used, human handoff, and any tool failures. If a task needs a different reasoning effort or a narrower brief, make that a documented change rather than attributing every difference to the model.

Make the cutoff observable

Before 14 October, list every active task or configuration that still chooses gpt-5.5, update the ones with a supported replacement, and run the important paths. For scheduled work, check that the next scheduled run starts and produces the expected artifact. Keep a note of settings you could not update because of workspace access or rollout. After the cutoff, inspect failures and remove stale selections instead of assuming that a client silently chose the right model.

This is a Codex and ChatGPT sign-in migration. API-key users should check their own API model availability and pricing separately; OpenAI explicitly says the GPT-5.5 retirement announcement does not apply to the API. For choosing between the new Codex models by task, see the Sol/Luna guide. For a measurable pilot, use the Codex credit and review-time worksheet.

Editorial note: This checklist is based on OpenAI's GPT-5.5 retirement and workspace guidance and model selection documentation, checked 24 September 2026. The AI Take has not tested every plan or workspace rollout.