GPT-5 Is Overtrained Roughly 100x Past Chinchilla Optimal

追加: 2026-05-28

[00:00:00]Open AI also realizes that in their surveying a certain amount of tokens per second that tells you how much data went into the free training of GPT-5. It even if it's like 50% off or something, that is that is sort of wild that you can sort of first principles these kinds of numbers.

[00:00:16]>> also I mean this is why you should just like approximate every way because like there's so big error bars on this. But yeah, I know it's kind of like empowering to just like set A equal to B and figure it out. Yeah, yeah. That's super cool.

[00:00:26]Okay, so in this sort of trying to deduce things we can publicly look up the prices of the APIs of these models and maybe you can learn something from that.

[00:00:37]So first with longer context >> [clears throat] >> Gemini 3.1 is um 50% more expensive if you go over 200k tokens than if you're below 200k tokens.

[00:00:53]I mean at a high level I understand why that might that be, but why specifically 50%?