Since last week it feels Opus 4.6 is producing only ■■■■■■ code and is from time to time ignoring Skills and Rules. The code is very buggy, plans are not executed to end and there is a lot of halluzinations I’m not used to have before up until about mid last week. You start explaining in great detail what you expect the agents to do and in most times only 20% is actually done. Is it just me or do others experience same issues ? I’m on almost most recent Cursor Version (did not install todays push).
Hi @Peter_Jaeger ,
It’s not broken; however, there are some recommendations I can make if you’re not seeing the output you’re looking for. You can up the reasoning level to Max from the default of High.
This thread gets very into the weeds, but if you’d like insight into Claude’s API changes I would check this thread: [MODEL] Claude Code is unusable for complex engineering tasks with the Feb updates · Issue #42796 · anthropics/claude-code · GitHub
@kevinn I’m well aware of the Feb changes - and mitigated them but what’s concerning is that since a week Opus got really unusable. I actually switched to Codex late last night and got the problems solved - almost immeaditly. So I would cnsider Opus broken for now - which is sad.
Sorry about the frustration, Peter. I’m glad you found a good workaround solution.