I asked Claude Code to write what it learnt to a file and moved to do something else. When I came back, it not only failed to do it but also consumed 32USD.
https://npub1nccwjspr3nv7h67xx2qhdh2dzzvp...
I'm sure Claude algo is pushing to consume more when billed per token. When I was using OpenClaw with it, first ai used by token, it was burning 5USD in an hour doing nothing special. Then I linked it to a Claude Pro at 20USD and I never hit any period limit. But now they blocked it and force to bill per token again, and it's burning credit like hell again. Other models, even counting for price difference are burning way less token, but then I have to retriy some tasks multiple time to achieve to a solution, so not sure what is best in the end.