GPT-5.5 Retires From ChatGPT and Codex on October 14: Migration Checklist

GPT-5.5 is scheduled to leave ChatGPT, ChatGPT Work and Codex on October 14, 2026. The API is not part of this retirement. That split matters because a team can migrate its API workloads correctly and still discover that saved workspace defaults, custom agents or scheduled Codex tasks point to a model that is no longer available.

What is changing on October 14

OpenAI’s September 14 retirement notice schedules GPT-5.5 removal from ChatGPT, ChatGPT Work and ChatGPT-authenticated Codex across all plans for October 14. The API is unaffected. Current migration guidance, checked October 12, supersedes the earlier replacement advice: Plus, Pro, Business, Enterprise and Edu users should select gpt-6-sol when available; Free and Go users should select gpt-6-luna in the desktop app when available.

This is a product-surface retirement, not a claim that every endpoint named GPT-5.5 disappears everywhere. The migration scope therefore depends on where a workflow runs and how it authenticates.

The four surfaces to inspect

1. Personal ChatGPT settings: saved model choices, custom GPT instructions and repeated manual workflows.
2. ChatGPT Work: shared workspace defaults, internal templates and guidance that tells employees to select GPT-5.5.
3. Codex: local configuration, managed settings, custom agents, saved tasks and scheduled work authenticated through ChatGPT.
4. API systems: document these separately. The retirement notice says they are unaffected, so changing a stable API integration only because the ChatGPT product is changing could create unnecessary risk.

A migration sequence that preserves evidence

Start with discovery, not replacement. Search configuration files, task definitions and internal documentation for both the display name and model identifier. Add an owner, execution surface and last-known successful run to the inventory. That turns a vague model change into a finite queue.

Next, select a small acceptance set for each workflow. A content brief might need source-link preservation, heading quality and word-count checks. A coding task may need tests, linting and diff review. A reporting task may need stable field names and totals. The point is not to prove that two models are identical. It is to identify which outputs are operationally important.

Change one surface at a time, record the result and keep an available fallback or a safe pause mode until the replacement is accepted. For scheduled work, confirm that the schedule actually fires after the change. A configuration file that parses successfully is not the same as a completed run.

What this means for SEO and publishing teams

Most SEO risk is indirect. A model retirement does not change Google’s ranking systems, but it can break the workflows that create briefs, refresh reports, monitor pages or prepare internal links. The failure mode is often silent: a scheduled task stops, a custom agent falls back to another model or a team member improvises a different process.

Publishers should preserve the human gates that matter regardless of model: verify primary sources, distinguish confirmed facts from inference, review links, check dates and inspect the rendered page. Our metric-contract approach to AI-assisted reporting is useful here because it defines the output before the model changes.

Use a migration register, not a memory test

The downloadable register includes execution surface, authentication method, current model, target model, owner, acceptance check, last successful run and rollback notes. It is intentionally plain CSV so it can be opened in a spreadsheet, committed beside configuration or imported into a task tracker.

Download the GPT-5.5 migration register

What not to infer from this notice

• It does not say GPT-5.5 API access ends on October 14.
• It does not establish that a replacement produces identical outputs.
• It does not mean every workflow should be migrated to the same model without testing.
• It does not change search rankings or AI citations by itself.

Before the October 14 deadline

Inventory the remaining GPT-5.5 selections and assign owners first. Verify that the replacement is available for the actual plan, client and workspace, then run each workflow’s acceptance set. Update defaults and documentation, confirm the next scheduled run, and escalate unresolved access or output failures before the deadline. In managed workspaces, administrator enablement may be required. Do not retain the retiring model as the post-deadline fallback.

Primary source

OpenAI’s retirement notice and current model guidance, checked October 12, 2026. This revision corrects the replacement targets and removes the stale seven-day countdown; it does not report a completed migration test.