I've seen big drops in Opus 5.5 (on high/x-high) & Fable. Both in terms of following instructions / workflows execution and overall code quality. Also the token burn-rate increased a lot, I've reached limits of 3 Ultra subscriptions during last week & had to spend 1.5k on API usage during the weekend just to be frustrated with the stupidity of the output.
Weirdly enough, Sonnet quality seems unchanged.
> & had to spend 1.5k on API usage
You rewarded Anthropic's failure with more dollar spend. I get why you did it but...