[jackie293746]

Claude Opus 4.6 regularly makes up shit and hallucinates. I'm not a detractor by any means but "exceptionally rare" is fantasyland.

[thrawa8387336]

Can vouch for this, plus, when it does work, stuff can take forever. Then, if I let it unsupervised, higher risk of doing the wrong thing. If I supervise it, then I become agent nanny.

[surgical_fire]

I have been experiencing it too.

I honestly am finding Codex considerably better, as much as I despise OpenAI.