Hacker News Reader

Hacker News

Comment by SV_BubbleTime | Hacker News Reader

Parent Full thread

SV_BubbleTime

> Pretty easy to hit $100 an hour

I don’t see how that can be true, but if it is…

Either you, or I are definitely use Claude Code incorrectly.

View on HN

It’s definitely easy with an API key I hit 200$ in an evening. I didn’t think that could be possible. Horrifying

To be clear, this is a lot of full-scale reading and (re)writing, without any rules, promots, "agents"/code to limit your resource usage, right?

Nobody's asking for $200 in single-line diffs in less than a day - right?

Right, this is having Claude Code just running as an agent doing a lot of stuff. Also tool use is a big context hog here.

It’s not about a single line diff but the same prompts with cursor does not end costing that much

Yes - do you understand why?

Yes I do, it’s just for new comers who are used to cursor where, without careful prompting you just lock yourself out of premium requests, that’s not immediately a given that CC is more dangerous and does not work the same way at all. Of course it requires careful planning but the trap is easy to fall into.

In my experience, it's more about the tool's local indexing and aggressive automatic upload and model usage limitations to avoid them (and you) overpaying.

People are recreating this with local toolchains now.

This is around what what Cursor was costing me with Claude 4 Opus before I switched to Claude Code. Sonnet works fine for some things, but for some projects it spews unusable garbage, unless the specification is so detailed that it's almost the implementation already.

> unless the specification is so detailed that it's almost the implementation already.

You mean… it’s almost exactly like working with interns and jr developers? ;)

Yeah, it's like an idiot savant intern who has memorized all the API docs in the world, but still somehow struggles to work independently.

And never remembers what we talked about 5 minutes ago on a different thread.

This is where something like Perplexity's "memory" feature is really great. It treats other threads similarly to web resources.

I would love to understand better just how Perplexity is able to integrate up-to-date sources like other theads (and presumably recent web searches, but I haven't verified this, they could be just from the latest model) into it's query responses. It feels seamless.

Reply on news.ycombinator.com