Brundage argues the CEO was likely guessing rather than leaking
it's more than 2x the total number of params of glm 5.2, and recently cursor CEO said opus/5.5 were 1.5T models (not sure what to think about it), but yeah i think it's still relatively small and i expect larger models to be served in production with newer hardware (if they don't get restricted)
@eliebakouch Isn’t that small though then? Would have guessed things were much bigger now
Users affirm CEO insights on high margins for large closed AI models as likely true and rational given expected knowledge of model sizes exceeding 1.5 trillion parameters.
No Digg Deeper questions have been answered for this story yet.
@Miles_Brundage i would expect him to know this information tbh so even if this means the margins are crazy i think the most rational thing is to assume it's true aha (see
i have a hard time believing this. at the same time it would make sense that the CEO of cursor know the size of closed models, but this would mean the margin on these models is CRAZY compared to oss models.
this plot doesn't even account for the fact that serving at scale reduces cost (much higher batch, better hardware) and that they have many more engineers working on optimizing the inference stack ect.. (this is a log-log plot btw)
@eliebakouch Yeah, I saw the Cursor thing but discounted it bc it seems like the kind of thing he would not be allowed to say so it must have been a guess, but dunno…
it's more than 2x the total number of params of glm 5.2, and recently cursor CEO said opus/5.5 were 1.5T models (not sure what to think about it), but yeah i think it's still relatively small and i expect larger models to be served in production with newer hardware (if they don't get restricted)
@eliebakouch (Also possible he wasn’t allowed but also it’s true… anyway)