@thisisdimm they probably using flash model to understand what to do for their flagship model (3.8 flash rumored already exist). Ever since their restructuring i think this made sense.
@keywordian I already activate one since 2 years ago. It is quite good.
However, I heard stories they turned it off for "not used" one without notice. Never experienced it myself
@thdxr TPS is linear by time, and TTFT is constant by time. You can't add them together. You can make other measurement like Time per Task (probably what you want) or Time to 1k Token etc.
@xxJayyyyy they are not making it because of the customer. they are making it because it is dispersing compute usage. Its kind of obvious, the new user one is just PR
@tylertracy321 Some (more modern) elevator do allow it. I think most people assume it doesn't. Like 60% of elevator I come across allow you press (or double press) to cancel one
@PawelHuryn Gemini 3.7 flash isn't competitive nor valuable at API pricing IMHO. I think the google subs is genuinely the only thing make it worth it (it is really generous). It also behave weirdly, in my testing it can do something really good but really bad at other things
@KeroBasalos@karanC_12@ilyasut I doubt it, They have major outage the other day. And they're lacking the compute to basically giving 1 Qd, + 100T token.
@nexxeln Proof it: by creating a sentence / paragraph. Then Sha256 it.
Example "The actual m0del is X, see I actually know -nexxel" is ....963812f993c5aa1907218722c463e0153dc6639e6f561845c0f
So we know you're not lying when it's actually released, and you don't need to leak it now
@poteto@cherry_mx_reds is the cancelling sub to get a sub still available? just cancelled claude because they ambiguous ending of 50% limit. would love to try cursor