Playbooks · Coding tasks
GPT-5 snapshots and o3 shut down Dec 11
Written November 12, 2026
If your code names a GPT-5 snapshot or o3, it stops working on December 11, 2026. Six weeks later the same thing happens to GPT-4o and GPT-4.1 — and to every fine-tuned model built on any of them. OpenAI's deprecations page names a replacement for each row, so the move is a model-string change for most teams, and a rebuild for anyone who fine-tuned.
The December 11 list
OpenAI's deprecations page announced these removals on June 11, 2026. All six rows shut down on December 11, 2026:
gpt-5and itsgpt-5-2025-08-07snapshot move togpt-5.6-sol.gpt-5-minimoves togpt-5.6-terra, andgpt-5-nanotogpt-5.6-luna.gpt-5-promoves togpt-5.6-solwithreasoning.mode: pro.o3moves togpt-5.6-sol, ando3-protogpt-5.6-solwithreasoning.mode: pro.
Fine-tuned versions of these models follow the same schedule as the base model under them. A fine-tune of gpt-5 is a December 11 problem, not a separate one.
January 14, 2027: the GPT-4 rows, fine-tunes included
Announced on October 14, 2026, the GPT-4 series rows shut down on January 14, 2027: GPT-4o (the gpt-4o-2024-08-06 and gpt-4o-2024-11-20 snapshots), GPT-4o mini, GPT-4.1 and GPT-4.1 mini. OpenAI names gpt-6-luna as the replacement for every one of them. The same sentence about fine-tuned versions applies here too, and this is where it bites hardest: a custom model trained on GPT-4o last year has no home after January 14 unless it is rebuilt on a newer base.
The wider fine-tuning clock is on the same page. From January 6, 2027, active existing customers can no longer create new fine-tuning jobs at all, and inference on a fine-tuned model is switched off when its base model is deprecated. So the practical order is: inventory your fine-tunes first, then your plain model strings.
Let a script watch the dates
OpenAI gives you a way to catch the next round without reading the page. The List models endpoint returns an optional shutdown_date field for each model, as a YYYY-MM-DD string or null when no shutdown has been announced. A small scheduled job that lists your models and flags any with a date attached turns deprecations like these into a ticket months ahead, instead of an outage on the day.
Limits
- These are API shutdowns. They say nothing about model names inside ChatGPT, Cursor or any other product built on the API; each product makes its own choices.
- The replacements above are OpenAI's recommendations from the deprecations page. Whether
gpt-5.6-solbehaves like your tunedgpt-5snapshot on your workload is a test only you can run, and December 11 is close. - OpenAI's stated minimum notice is six months for generally available models, which both of these rounds meet. Preview models can go with far less notice, so the endpoint check matters more for those.
Related playbooks
- GPT Image 2 shuts down Feb 10, 2027
- GPT-Image-2.6: Flare or Sunburst?
- GPT-6 Cyber: OpenAI's restricted security model
- GPT-6.1 Sol, Ultrafast, and the new Pro tiers
Sources
Back to: Coding tasks playbooks hub · Also: GPT Image 2 shuts down Feb 10, 2027 · Gemini Embedding 2: update your model string