Their 20$ tier currently isn't serving their best model, and they insulted their users by putting out an ill tested opus 5.0, which is the worst experience ive personally had using a model in probably 2 years(obviously adjusting for expectations at the time of release).
People were very skeptical about how much investment most companies put into hardware/data centers two years ago, and anthropic was more conservative than OpenAI here, so it's potentially hurting them now.
(Opus is a separate story: it does seem to have improved in coding in my experience, most weirdness seems to be its human communication)