Anthropic's issues, as I see them, are:

1. They acted as if they were so far ahead capability wise that they could stop listening to their users.

That is basically it. It is an extremely common belief that Opus 5 acts like a condescending wannabe-thought-leader, yet Anthropic's response to this is largely been "You're using it wrong. Try deleting all your config files".

They released a "concise" output format, but pretty much swept it under the rug, despite it being one of the loudest complaints about their models. It also, generally, does not work as described.

Not only did Opus 5 get much more difficult to work with, but it got, from a customer view, significantly slower. The token rate might be the same, but if every interaction takes 50% more tokens, it's 50% slower.

All this time OpenAI has released a slew of new models, lowered the price on them (which, bluntly, 80% of user count don't actually care about since they're on subscriptions), increased subscription capacity, and increased response speed.

It's not about cost. It's about Anthropic being the frontier-lab version of the marathon runner who decides to celebrate to early, and then loses the race.