ChatGPT✓ ReviewedTest carefullyAnnounced 21 September 2026Reviewed 21 September 2026OpenAI is retiring self-serve fine-tuning, and a lot of its own tooling with it.
OpenAI has published a timetable that closes self-serve fine-tuning to everyone by early 2027, alongside shutdown dates for its Evals platform, Agent Builder and reusable prompts. The same register sets out how much warning any model gets before it disappears.
Bit’s takeaway
What changed
OpenAI's deprecation register now carries a wind-down schedule for self-serve fine-tuning: new organisations lost the ability to create fine-tuning jobs on 7 May 2026, organisations without recent inference were disabled on 2 July 2026, and active customers stop being able to create new jobs on 6 January 2027, with inference disabled once the base model is deprecated. Separately, the Evals platform, Agent Builder and reusable prompts are all listed for shutdown on 30 November 2026, and the Assistants API shut down on 26 August 2026 after a year of notice. The register also states the notice OpenAI commits to: at least six months for generally available models, at least three months for specialised variants, and as little as two weeks for preview models.
Why it matters
Most people never think about the model underneath the tool until it changes. When a model is retired the usual experience is not an error message but a quietly different answer to a request that worked last month, which means you tend to hear about it from a colleague or a customer rather than an alert. A published timetable is the difference between a scheduled task and a bad Monday.
Who should care
- Anyone who has built saved prompts, assistants or routines on a specific model
- Teams running anything customer facing on an AI provider's API
What to do
List what you actually depend on and which model each thing uses, then keep one example of good output per task so you have something to compare against. Test the replacement well before the cutoff rather than on the day, and check the retirement date for the platform you run on, because partner platforms set their own schedules.
The human take
Tools change fast. Your judgment matters more.
A person decides which prompts and routines are worth testing, judges whether the replacement output is still right for the job, and picks the day to move. The provider supplies the deadline, not the judgement about what still works.
Turn this into a workflow
Do not just read the change. Try it safely.
Bit starts from this reviewed update and helps you build a safe way to use it on your own work: what AI can prepare, what you still check, and where the workflow should stop.
Turn this into a safe workflow with BitSources and method