i've spent around $250 on anthropic at this point. keep in mind this is all api pricing.
no subscription, just api
i never actually had a claude subscription. it was all pay-as-you-go api credits. i'd top up, use it for a while, top up again. it worked fine but i was always watching the balance.
the opus problem
the thing is, opus was actually really good. like, noticeably better. but i basically couldn't use it because it would just burn through my credits instantly. so i ended up on sonnet and haiku most of the time, which were fine, but i knew i was using the worse models because of cost.
every request was awful, how many tokens are in my window. since on a new session these 10 bucks would last a few minutes but once i was like 2 prompts in they were instantly gone.
switched to codex
a while back i got the codex subscription. $20 a month and i get like $300 of api usage. i basically never hit the limit.
now i just use gpt-5.5 on low reasoning for everything. it's fast, it's good enough, and i don't think about cost at all. even if i used api billing for it it's much cheaper than opus.
the backup plan
worst case if i do hit limits on codex, i have the opencode go subscription. the openweight models are actually really good now too. i just use kimi k2.5 and k2.6. i do really enjoy them though.
bottom line
i've moved away from claude and anthropic, firstly because the limits in the codex subscription are so much better, secondly because i now have 2 subscriptions and don't need claude anymore, for most things i have codex for frontend kimi and lastly because of all the shit they're pulling these days.