News for hackers. Stuff that matters.
Re:OpenAI's Unit Economics by Lerc on Thursday January 29, 2026 @11:54AM (#46812798)
They do still train models from scratch, and they are still making larger models.

You would expect that to use a lot more resources.

It seems to be that GPT 5.x are all likely to be extensions of the gpt5 base model with similar numbers of parameters.

The money spent on extending base models would be dwarfed by the scale increase of the next major number.

There are no comments on this comment yet.