I got curious and tried out the tool because you can use it with a custom provider. After a little bit of testing it seems to work well, but holy does using it burn through tokens fast. With just a few queries asking it about a repo to test it out, nothing even requiring it to write, compile, or test code I burned through 50 cents using Sonnet 4.

I could easily burn through a minimum of $20 in a day.

I'll probably have to stick with Aider for now. I still like reading the patches the LLM is applying and keeping myself more in the loop.

nostr:nevent1qvzqqqqqqypzpq35r7yzkm4te5460u00jz4djcw0qa90zku7739qn7wj4ralhe4zqqsrc480tn28jcr43mx494k6lm9d348e03ajhz6rwld7sz2r2lsy39c73fzp4

Reply to this note

Please Login to reply.

Discussion

Yeah Claude code is also expensive as fuck

No kidding, what I didn’t realize when I gave it a go is that it’s not just an agent calling tools in the background, it’s actually an agent calling agents calling tools. One agent instance can start to get a little pricy, but trying to use an agent firing up other agents all day gets really expensive.

I do wonder tho how much of the design decisions are driven by the fact that the more you use their API, the more money Anthropic makes.