OpenAI's Nov 30 and Dec 11 cutovers: a migration checklist.
OpenAI is retiring 13 tools and models in the next 66 days. OpenAI Agent Builder, OpenAI Evals and reusable prompts shut down on Nov 30. GPT-5, o3 and o3-pro leave the OpenAI API on Dec 11, and GPT-4o and GPT-4.1 follow on Jan 14. Every one has a named replacement. The risk is not the swap. It is switching without testing.
The signal
OpenAI's API deprecations page lists three dates that matter to any team building on the OpenAI API or ChatGPT. On Nov 30, 2026, OpenAI Agent Builder, the OpenAI Evals dashboard and API, and the v1 / reusable prompts API shut down. On Dec 11, gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro are removed. On Jan 14, 2027, gpt-4o, gpt-4o-mini, gpt-4.1 and gpt-4.1-mini go too. That is 13 tools and models in 66 days. Source: OpenAI API deprecations
Why it matters
Four of five layers in a typical agent change. The interface (your app or ChatKit) stays. Orchestration built in Agent Builder, prompt objects, the model, and the evals that check all of them do not.
Nov 30 matters more than Dec 11. Losing a model is a swap. Losing your evals is losing the proof that the swap worked. OpenAI Evals have been read-only since Oct 31 and disappear on Nov 30. Export them first.
Every retirement has a named way out
OpenAI publishes a replacement for each item: Agent Builder workflows move to the OpenAI Agents SDK or ChatGPT Workspace Agents; evals move to Promptfoo or your own harness; prompt objects move into your application code; the GPT-5 family moves to gpt-5.6 (sol, terra, luna); gpt-5-pro and o3-pro map to gpt-5.6-sol with pro reasoning mode; GPT-4o and GPT-4.1 move to gpt-6-luna. ChatKit stays.
The Enso take
A new OpenAI model is not a drop-in swap. Output format, tool-calling behaviour and cost all move between model families. The safe migration is a loop, not a swap: find what is pinned, rebuild flows and move evals in parallel, test the replacement against your own evals, switch before the deadline, then keep re-testing every time OpenAI ships a model change.
What this means for a CTO or a ChatGPT Enterprise admin: you have three weeks to test replacements, not one night to swap a model name.
The Monday move
1. Search for pinned models. Grep code and configs for gpt-5, o3, gpt-4o and gpt-4.1.
2. List every Agent Builder workflow. Rebuild each one on the OpenAI Agents SDK or as a ChatGPT Workspace Agent.
3. Export your evals. They have been read-only since Oct 31 and disappear Nov 30.
4. Test the replacement first. Run those evals on gpt-5.6 or gpt-6-luna, then switch.
As an OpenAI Select Partner, Enso Labs builds and runs agents on OpenAI models as a managed service, with evals that catch a model change before your users do.
Powered by Enso Labs
Frequently Asked Questions
When does OpenAI Agent Builder shut down?
November 30, 2026. OpenAI Evals (dashboard and API) and the v1 / reusable prompts API shut down the same day. ChatKit stays available. Source: the OpenAI API deprecations page.
What replaces OpenAI Agent Builder?
OpenAI names two paths: the OpenAI Agents SDK, or ChatGPT Workspace Agents. Workflows built in Agent Builder need to be rebuilt on one of them before November 30.
When are GPT-5 and o3 removed from the OpenAI API?
December 11, 2026: gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro. OpenAI's replacements are the gpt-5.6 family (sol, terra, luna); gpt-5-pro and o3-pro map to gpt-5.6-sol with pro reasoning mode.
What happens to GPT-4o and GPT-4.1?
They are removed on January 14, 2027 (gpt-4o, gpt-4o-mini, gpt-4.1, gpt-4.1-mini), replaced by gpt-6-luna.
What is the first thing to do?
Export your evals before November 30. They have been read-only since October 31, and they are the only way to prove a replacement model works for your use case before you switch.
Operate — This piece is part of Enso's Operate practice.
Ready to run this in production? In a 30-min call we'll review your stack and design a retainer that keeps it healthy.