I have really been struggling with Opus 4.6 the last few days, where it has become (even more) lazy, and just doesn’t do a good job at performing or finishing a task. It started telling me “open the document and make the following changes” instead of just doing that, which is what I am used to doing. It consistently repeats the same errors: when told “this is wrong because [xyz], fix it” it will just apologise, and then make the same or a similar error for a few turns. Until I shout at it and tell it to not be lazy, go deepr, think harder, whatever, it will just do an incredibly poor job. This is across sessions, different chats, different subjects, and hte last few days it has been even worse than we have seen the last few weeks prior to the Opus 4.7 release.
All of this eats tokens at an increasing rate as the chat and the frustration grows. More stupid answers is more tokens, is more money for Anthropic. Coupled with yet another tweak to the usage limits that was foisted on us yesterday, it has become intolerable.
At this point I am done with throwing money into a crappy tool that once used to be excellent. I already moved from Max to pro, and now I am cancelling. ChatGPT is not as good, but at least it is reliable in its’ “almost the best” quality for many tasks.
For coding I have been on Kimi K2.5, GEMMA 4 locally, and ChatGPT for a few weeks now, and my spend has dropped dramatically without a meaningful drop in quality, but with a reasonable drop in speed.
It is tragic how Anthropic ruined a good thing, and fumbled so much goodwill. I am in Enterprise technology, and design and build large-scale solutions. I used to recommend. Claude, but I can no longer do so.