Deprecation of fine tuned models, but still can't access newer ones?

API Deprecations
,
1

The latest deprecation announcement, makes it sound like several models, like ft-gpt-4.1-nano-2025-04-14 are being shut down. In that particular example, it says to use gpt-5-nano instead. But I have no ability to fine tune gpt-5-nano, mini, or regular gpt-5 still. Will access to this be rolling out soon? Or am I misreading the deprecation announcement? Or are you actually removing fine tuning support?

1 Like
2

Good question!

I have pinged the team for clarification.

3

Yeah, what about fine tuned model? Does our fined tuned model also be deprecated too? Or is it just the base model that get shutdown?

4

Hello,

You’re not misreading the shutdown notice: fine-tuned versions of gpt-4.1-nano-2025-04-14 are scheduled to be removed on October 23, 2026, and the deprecations page lists gpt-5-nano as the recommended substitute.

The “substitute model” column is a migration target for using the API after shutdown; it does not currently mean that gpt-5-nano is available for fine-tuning. Fine-tuning itself is not being removed, but GPT-5-family models are not currently listed in the public fine-tuning docs as supported fine-tuning base models.

If you need a fine-tuned replacement today, please use one of the currently supported fine-tuning models documented in the fine-tuning guides. If your goal is specifically to migrate from a fine-tuned gpt-4.1-nano model, we recommend evaluating base gpt-5-nano against your existing evals/prompts, since that is the documented substitute for that deprecated model family.

Reference:

5

Well, this is very uncool. We’re using nano specifically because the response times are much better. gpt-5-nano absolutely will not do our classification correctly without fine tuning.

We all know that gpt-5 fine tuning exists and a bunch of people have access to it already (people have published papers!). Why can’t the rest of us start fine tuning on gpt-5 (nanon/mini/regular)?