I'm not. Yes we normalized 1m context window and models tend to hallucinate less.

But models have been somewhat stagnant since Opus 4.6/7.

And in some regards there were even regressions like Claudeisms that are load bearing.