undefined | Better HN

0 pointsrubicon332mo ago0 comments

But actual progress seems to be slower. These modes are releasing more often but aren’t big leaps.

0 comments

We used to get one annual release which was 2x as good, now we get quarterly releases which are 25% better. So annually, we’re now at 2.4x better.

wahnfrieden2mo ago

GPT 5.3 (/Codex) was a huge leap over 5.2 for coding

rubicon33OP2mo ago

Eh, sure, but marginally better if not the same as Claude 4.6, which itself was a small bump over Claud 4.5

minimaxir2mo ago

Due to the increasing difficulty of scaling up training, it appears the gains are instead being achieved through better model training which appears to be working well for everyone.

j / k navigate · click thread line to collapse