both opus4.6 and gpt5.3 are somehow really slow. like yeah the inference itself is def much faster, but they both just keep stuffing more and more shit into their context even when it's really not necessary. and only start actually writing code when they get to like 100k tokens worth of context