OpenAI has officially retired GPT-5.5 in Codex for users accessing Codex through ChatGPT accounts, implementing a planned sunset that took effect on October 14 at 12:00 PM PDT (19:00 UTC). OpenAI had announced the October 14 retirement across ChatGPT, Work, and Codex on September 14.
The change does not apply universally across OpenAI's developer catalog. Users calling models through the standard OpenAI API and those authenticating Codex environments directly with an API key remain unaffected by the milestone.
Replacement Models by Account Tier
For accounts that authenticate through ChatGPT, OpenAI documentation directs users to different successor models depending on subscription tier and client environment:
- Plus, Pro, Business, Enterprise, and Edu: Users are instructed to select GPT-6.1 Sol (
gpt-6.1-sol) where available. - Free and Go: Users operating in the desktop application are instructed to select GPT-6 Luna (
gpt-6-luna) where available.
Availability varies across clients and plans. OpenAI notes that selecting a replacement identifier or updating a default setting does not inherently confer model access to accounts that lack permissions. In Enterprise environments, workspace administrators must explicitly enable the replacement models before affected team members can use them. Transitioning models does not modify existing workspace permissions or sandbox configurations.
Required Configuration Updates
OpenAI instructs developers to replace saved gpt-5.5 selections with a model available to their account and client.
Teams should inspect configurations where gpt-5.5 may be hardcoded or set as an active default:
- Workspace defaults and saved preferences
- Centrally managed configurations
- Custom agents and workflow definitions
- Scheduled background tasks and operational scripts
- Command-line selections, using
/modelin an interactive session or--modelwhen starting Codex
Model assignments must be confirmed individually across client surfaces before relying on them in scheduled or interactive work.