[rubicon33]

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

[gallerdude]

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.

[wahnfrieden]

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

[rubicon33]

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

[minimaxir]

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.