I’ve been using Opus 4.5 intensively over the last week or so during the discounted pricing period, so feel that I’ve acquired a good sense of the characteristics of this extremely impressive model. I also have a lot of Sonnet experience. Up until yesterday, Opus was diligent, thoughtful, and put in the extra effort to add polish and nice features to solutions.
It’s just a vibe, but as of this morning, the model feels… different. More recalcitrant, more terse, prone to making syntax errors, less hard working, more eager to defer complex work to later sessions or successors where it would have just powered through each request sequentially.
There’s nothing concrete to justify this, and I know that Cursor and Anthropic have denied ‘nerfing’ models in response to demand. I am curious if other Cursor users are having a similar experience today.