Hacker News Reader

Comment by strangescript | Hacker News Reader

I have been using Claude Code, and its a bit pricey, but has exceeded my expectations so far.

Do tell. How is it better than Cursor, Windsurf or (probably closest) Aider? Does it do anything the former don't, that was not possible before? I did not really understand the value prop, but am super interested to learn more.

I don't like Cursor, because I like PyCharm as my very awesome IDE. Claude Code is a terminal app, it's something that I just issue commands to and it mostly works, and then I check everything in PyCharm.

I tried Aider a couple of months ago and it gave me crappy code. Maybe it was the model I used back then, but I didn't like it. I had much better experience with Claude Code.

There are valid criticisms of Aider (I feel like it might use more tokens than it needs to, but I haven't done a lot of comparisons), but output quality is certainly the model.

Are you running it from the built-in terminal in PyCharm? I also use that IDE and want to give this a swing.

Can you elaborate on "check everything in pycharm?"

Cursor's in-house models are vastly inferior to GPT or Claude. I bet the majority of Cursor users are still using one of these two under the hood.

Yes, but the question was about Claude Code and not the underlying model.

Before Claude Code, out of the alternatives you mentioned, I'd only tried Windsurf. I find Claude Code to be roughly comparable to Windsurf in terms of capabilities, but I can use it from my terminal. I'm much more at home in a terminal than I am in VSCode (which Windsurf is a fork of).

Oh wow you owe it to yourself to try aider then. Architect mode especially, where you use one (usually reasoning) model to plan and one (usually Claude) to write code, is pretty awesome.

Nope -- he's giving us his personal feedback and experience. I agree with him. I dislike Cursor and Windsurf and enjoy Claude Code alot.

Bother to try it for yourself, and make your own decision. I know it make actually be a bit of work...but you'll probably find everyone has their own style from the web API, to some IDE. No need to be so hostile.

I do not read hostility there, just directness. He has a valid inquiry and states he is interested to learn more. Assume good faith.

Agree. Not sure how, “I’m super interested to learn more,” indicates hostility.

I think ranscendental50's reply itself was unnecessarily hostile and don't agree with the way they responded, but I can see why they assumed hostility. "Do tell" is unfortunately to a large extent used passive aggressively on the Internet, and generally actually means "Don't bother telling me, you're obviously ignorant".

This seems to be a case where it's meant genuinely, but it's not surprising that not everyone picked up on that.

Claude Code is amazing. Typically it has costed me about $25/day of using it but it's well worth it. I ran into the $100/month limit and now waiting for my account upgrade so I can use it some more.

Wow, how are you using so much? Are you basically prompting it to build out a whole application?

I built a large front-end component with it.

Does Claude Code charge based on API usage?

And does it let me limit the files it's sending?

Yes, by API usage. I think you can’t limit which files it sends.

They’ve made a recent change to Claude Code that apparently makes it respect your gitignore

https://github.com/anthropics/claude-code/issues/79

In that sense, Aider is definitely better.

A coworker was trying both out on a project, and was finding Claude Code to be 4x more expensive (both using the Sonnet-3.7 mode) - because he couldn't limit the context Code would send. As the project grows, that multiplication factor could be a lot larger.

Reply on news.ycombinator.com