When the base goes, the fine-tune goes

On 23 October 2026, OpenAI shut down a batch of older models, among them o1, o3-mini, o4-mini and gpt-4-turbo, and with them a fine-tuned version of o4-mini. On 14 October it gave the next date: gpt-4o, gpt-4o-mini, gpt-4.1 and gpt-4.1-mini go on 14 January 2027.1

Its deprecations page says it in one sentence: fine-tuned versions of these models follow the same shutdown schedule as their underlying base models.

The work you taught a rented model is only as permanent as the model you taught.

What a company actually loses

Not just a model name. A fine-tune holds months of decisions: which examples were chosen, which answers were corrected, what good looks like in your firm. When the base is retired, the weights you paid to adjust are gone, and the work has to be redone on whatever comes next, tested again and approved again.

Two more dates sit on the same page: existing customers can no longer start new self-serve fine-tuning jobs after 6 January 2027, and the vendor’s evaluation platform turns read-only on 31 October 2026.1

How owning changes it

• The weights are yours. An Eigna model is a file you hold, on the machine you choose. No vendor date applies to it.
• It stays the same until you decide. A new version goes live only when it beats the current one on your sealed test, and the previous one is one step back.
• The base is pinned. We record the exact open base model and its licence, so you can always rebuild what you own.

This is why the Model Deed, in development, names the owner, the release and the fingerprint of the weights. Ownership is a document, not an adjective, and a document does not have a shutdown date.