Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Claude Code on the web

anthropic.com

Post reply on HN
Print view View original

341–350 of 405 posts

Re: Claude Code on the web

#341

Post by deaux »

virtualritz wrote: post #336
mmaunder wrote: post #41

We were heavy users of Claude Code ($70K+ spend per year) and have almost completely switched to codex CLI. I'm doing massive lifts with it on software that would never before have been feasible for me personally, or any team I've ever run. I'll use Claude Code maybe once every two weeks as a second set of eyes to inspect code and document a bug, with mixed success. But my experience has been that initially Claude Co…

When you say Claude Code, what model do you refer to? CC with Opus still outperforms Codex (gpt-5-codex) for me for anything I do (Rust, computer graphics-related). However, Anthropic restricted Opus use for Max plan users 10 days or so ago severly (12-fold from 40h/week down to 5h week) [1]. Sonnet is a vastly inferioir model for my use cases (but still frequently writes better Rust code than Codex). So now I use Co…

Agreed that Opus is stronger than Sonnet 4.5 and GPT-5 High. It's the bitter pill - bigger, more expensive models are just "smarter", even if it doesn't always show in synthetic benchmarks. Similar with o1-pro (now almost a year old, an eternity in this space) vs GPT-5 high. There's also GPT-5 Pro now, which comes at an API cost of $120/M output, and is also noticeably smarter, just like Opus.

They all like to push synthetic benchmarks for marketing, but to me there's zero doubt that both Anthropic and OpenAI are well aware that they're not representative of logical thinking and creativity.

Re: Claude Code on the web

#342

Post by bapak »

pjmlp wrote: post #272
wahnfrieden wrote: post #58

Earlier quoted context omitted.

If your app runs on iPadOS, you already need to support every "screen size" (window size) Android is simply a much worse platform to make money on. Users spend <25% as much as iOS users. Why would they prioritize that?

Because 70% of the mobile phone world runs on Android. It is like trying to make a living selling games to macOS users.

Please take a look at the percentage of paying Android users. It just does not compare. It's useless to count 2 billion users in third world countries who never have and never will pay anything in-app.

Re: Claude Code on the web

#343

Post by ssk42 »

dns_snek wrote: post #305
hluska wrote: post #284

Earlier quoted context omitted.

Nobody has to give you examples. People can express opinions. If you disagree, that’s fine but requesting entire prompt and response sets is quite demanding. Who are you to be that demanding?

> Who are you to be that demanding? Let's call it the skeptical public? We've been listening to a group of people rave about how revolutionary these tools are, how they're able to perform senior level developer work, how good their code is, and how they're able to work autonomously through the use of sub-agents (i.e. vibe coding), without ever providing evidence that would support any of those grandiose claims. But t…

Context engineering is a critical part of being able to use the tool. And it's ok to not understand how to use a new tool. The different models combined with different stacks require different ways of grappling with the technology. And it all changes! It sucks that you've tried it for your stack (Elixir, whatever that is) in your way and it was disappointing.

To me, the tool inherently makes sense and vibes with my own personality. It allows me to write code that I would otherwise procrastinate on. It allows me to turn ideas into reality, so much faster.

Maybe you're just hyper focused on metrics? Productivity, especially when dealing with code, is hard to quanitfy. This is a new paradigm and so it's also hard to compare apples to oranges. Does this help?

Re: Claude Code on the web

#344

Post by raducu »

latexr wrote: post #322
raducu wrote: post #315

Earlier quoted context omitted.

> “this LLM tool I’m using now is the real deal". GPT-5 is not the final deal, but it's incredibly good as is at coding. Anecdotal, but it's something completely else in terms of capabilities, ignore it at your own peril, but I think it will profoundly change software development.

> ignore it at your own peril I’m not arguing for ignoring it, my point is different. > but I think it will profoundly change software development. The point is that this is said every time, together with “the previous thing to which the exact same praise was given , wasn’t it”. So it’s several rounds of “yes yes, the previous time the criticisms were right, but this time it’s different, trust me”. So everyone else i…

> So it’s several rounds of “yes yes, the previous time the criticisms were right, but this time it’s different, trust me”. So everyone else is justified in being skeptical.

True, but even the boy who cried wolf too many times eventually got his sheep eaten by the wolf.

I have my own personal anecdotal benchmarks and I never hyped LLMs before GPT-5.

Things that simply did not work before GPT-5 no matter how many shots I gave them, GPT-5 breezed through.

For me, it would take at least 2 generations of no felt progress in the models to call for diminishing returns, and I'm not seeing them.

Re: Claude Code on the web

#345

Post by pinkbanana21 »

mmaunder wrote: post #41

We were heavy users of Claude Code ($70K+ spend per year) and have almost completely switched to codex CLI. I'm doing massive lifts with it on software that would never before have been feasible for me personally, or any team I've ever run. I'll use Claude Code maybe once every two weeks as a second set of eyes to inspect code and document a bug, with mixed success. But my experience has been that initially Claude Co…

I found using z.ai a performant and cheap alternative for some tasks: https://z.ai/subscribe?ic=VWKNBI8LR8

Costs are 6x cheaper and it's way faster and good at test writing and tool calling. It some times can be a bit messy though so use Gemini or Claude or codex for that hard problems....

Re: Claude Code on the web

#346

Post by veidr »

swah wrote: post #177
mmaunder wrote: post #41

We were heavy users of Claude Code ($70K+ spend per year) and have almost completely switched to codex CLI. I'm doing massive lifts with it on software that would never before have been feasible for me personally, or any team I've ever run. I'll use Claude Code maybe once every two weeks as a second set of eyes to inspect code and document a bug, with mixed success. But my experience has been that initially Claude Co…

I haven't been able to get anything done with Codex. Claude Code is fast and "gets it". Also does better at running and testing its own stuff. Its very odd because I was hoping they were very on par.

Me too, but I know it's not just people shilling, or on the take, because a bunch of people I know personally have moved from Claude Code to Codex, and say it's better.

For me, though, it's not remotely close. Codex has fucked up 95% of the 50-or-so tasks I asked it to do, while Claude Code fucks up only maybe 60%.

I'm big on asking LLMs to do the first major step of something, and then coming back later, and if it looks like it kinda sucks, just Ctrl-C and git revert that container/folder. And I also explicitly set up "here are the commands you need to run to self-check your work" every time. (Which Codex somewhat weirdly sometimes ignores with the explicit (false) claim that it skipped that step because it wasn't requested... hmm.)

So, those kinds of workflow preferences might be a factor, but I haven't seen Codex ever be good yet, and I regret the time I invested trying it too early.

Re: Claude Code on the web

#347

Post by deaux »

extr wrote: post #116
PantaloonFlames wrote: post #97

Earlier quoted context omitted.

That’s a curious approach. Why would you use both? Why not just use the more reliable dependable option for all purposes?

Sonnet 4.5/CC is faster, more direct, and is generally better at following my intent rather than the letter of my prompt. A large chunk of my tasks are not "solve this concurrency bug" or "write this entire feature" but rather "CLI ops", merging commits, running a linter, deploying a service, etc. I almost use it like it was my shell. Also while not quite as smart, it's a better pair programmer. If I'm feeling out a…

Strangely enough this is one of the first times here I see someone with the exact same experience. GPT-5 is very prone to a style that would for most codebases be overengineering. I think as a large part of HN works on huge enterprise FAANG-like code, this is where it shines, so here it gets rave reviews of just being the best overall. But globally, for most developers, it's overengineering and adds a lot of unnecessary code to maintain. Sonnet in that sense remains "every man's coder". I've gone back from 4.5 to 4 now, having spent a good chunk of time with 4.5 it just seems like a slight overall regression with no real upsides besides being a little faster than 4.

Re: Claude Code on the web

#348

Post by deaux »

theshrike79 wrote: post #67
durron wrote: post #46

Earlier quoted context omitted.

Do you find this to still be true with the Sonnet 4.5 model?

I'm like 80% sure Sonnet 4.5 is just rebranded Opus. Sonnet 4 was a coding companion, I could see what it was doing and it did what I asked. Sonnet 4.5 is like Opus, it generates massive amounts of "helper scripts" and "bootstrap scripts" and all kinds of useless markdown documentation files even for the tinies PoC scripts.

It's very much not, so I'm more than happy to take that bet - how much are we wagering? Have you ever used each for non-coding tasks?

The generation of helper, markdown and bootstrap scripts are very dependent on your harness.

Re: Claude Code on the web

#349

Post by maherbeg »

fragmede wrote: post #156
maherbeg wrote: post #132

Earlier quoted context omitted.

lol yeah, but mostly just want to allow more types of reads for getting context, and primarily for test running / linting etc. I shouldn't have to approve every invocation of `pytest` or `bazel test`.

--dangerously-bypass-approvals-and-sandbox isn't enough for you?

I don't want unlimited writes. I basically want to unlock nearly everything but approve writes in some scenarios.

Re: Claude Code on the web

#350

Post by panarky »

latexr wrote: post #308
mmaunder wrote: post #41

We were heavy users of Claude Code ($70K+ spend per year) and have almost completely switched to codex CLI. I'm doing massive lifts with it on software that would never before have been feasible for me personally, or any team I've ever run. I'll use Claude Code maybe once every two weeks as a second set of eyes to inspect code and document a bug, with mixed success. But my experience has been that initially Claude Co…

Feels like with every announcement there’s the same comment: “this LLM tool I’m using now is the real deal, the thing I was using previously and spending stupid amounts of money on looked good but failed at XYZ, this new thing is where it’s at”. Rinse and repeat. Which means it wasn’t true any of the previous times, so why would it be true this time? It feels like an endless loop of the “friendship ended” meme with A…

Results > memes.

If we ship more for less because the new agent doesn't tap out, that's not a honeymoon, it's an upgrade.

Post reply on HN