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

Claude Sonnet 4.5

anthropic.com

Post reply on HN
Print view View original

481–490 of 819 posts

Re: Claude Sonnet 4.5

#481

Post by tclancy »

tclancy wrote: post #479
Aflynn50 wrote: post #374

When I see how much the latest models are capable of it makes me feel depressed. As well as potentially ruining my career in the next few years, its turning all the minutiae and specifics of writing clean code, that I've worked hard to learn over the past years, into irrelivent details. All the specifics I thought were so important are just implementation details of the prompt. Maybe I've got a fairly backwards view…

Eh, I’ve gotten over that. I’ve been using Claude recently on a personal project for a friend who wanted to take a known export file format and turn it into a list of good households for local political candidates to hit when knocking on doors. And I did that. But it’s been a while since I used pandas and numpy so I told Claude to swap out my loops for efficient code. And he did. Then, just for fun, I said, “Hey, sin…

That said, your man and I will have a talk about what a walkable route looks like to humans.

Re: Claude Sonnet 4.5

#482

Post by simonh »

hshshshshsh wrote: post #269
simonh wrote: post #239

Earlier quoted context omitted.

Not everyone hates their job and gets no satisfaction from it. Some of us relish doing something useful and getting paid for it.

Sure. I don't doubt it. But let's say if I can make a 100 million pounds appears on your bank account tomorrow. Will you say no to it and go back to your day job?

Both can be true. Being able to do better, more productively work can increase my quality of life. And yes, winning lottery millions would increase my quality of life even more.

However I don’t have lottery millions, but I do have a job and I would like to be able to do it better.

Re: Claude Sonnet 4.5

#483

Post by kshacker »

simonw wrote: post #204

I had access to a preview over the weekend, I published some notes here: https://simonwillison.net/2025/Sep/29/claude-sonnet-4-5/ It's very good - I think probably a tiny bit better than GPT-5-Codex, based on vibes more than a comprehensive comparison (there are plenty of benchmarks out there that attempt to be more methodical than vibes). It particularly shines when you try it on https://claude.ai/ using its brand n…

Sorry I downvoted this by mistake. Got confused between collapse and the down arrow (which I rarely use). Does someone know how to reverse a downvote. I do not see it.

Re: Claude Sonnet 4.5

#484

Post by FergusArgyll »

netdur wrote: post #355
peterdstallion wrote: post #157

I am a paying subscriber to Gemini, Claude and OpenAI. I don't know if it's me, but over the last few weeks I've got to the conclusion ChatGPT is very strongly leading the race. Every answer it gives me is better - it's more concise and more informative. I look forward to testing this further, but out of the few runs I just did after reading about this - it isn't looking much better

Yeah, ChatGPT is better overall, but with a twist, Gemini can actually be the best if you use AI Studio, tweak the config, and set up a good system prompt, kinda like how nano banana is SOTA, but Qwen-Edit feels more useful since it’s less censored, meanwhile ChatGPT is starting to feel slower and kinda showing its age, another example is Veo 3 being SOTA while the infamous Grok is technically worse but doing better,…

Gemini in AI Studio is so much better than in Gemini.com / app. You would think that signals they're going for devs over consumers, but they're a consumer company. A real head scratcher

Re: Claude Sonnet 4.5

#485

Post by colinplamondon »

Aflynn50 wrote: post #374

When I see how much the latest models are capable of it makes me feel depressed. As well as potentially ruining my career in the next few years, its turning all the minutiae and specifics of writing clean code, that I've worked hard to learn over the past years, into irrelivent details. All the specifics I thought were so important are just implementation details of the prompt. Maybe I've got a fairly backwards view…

That code taste of yours, your ability to understand the minutiae, makes you superhuman at managing agents.

Comments, docstrings, naming, patterns - by defining better approaches and hold agents to them, the results will be better. Way better.

You can't grow a meaningful codebase without solid underlying primitives. The entropy will eat you alive.

Systems architecture is becoming more important - systems that play well with agents wind up looking more like enterprise codebases.

Re: Claude Sonnet 4.5

#486

Post by nenenejej »

Mathiciann wrote: post #156
cloverich wrote: post #108

Please y'all, when you list supportive or critical complaints based on your actual work, include some specifics of the task and prompt. Like actual prompt, actual bugs, actual feature, etc. I've had great success with both ChatGPT and Claude for years, am around 3x sustained output increase in my professional work, and kicking off and finishing new side projects / features that I used to simply not ever finish. BUT t…

I am almost convinced your comment is parody but I am not entirely sure. You want proof for critical/supportive criticism? Then almost in the same sentence you make an insane claim without backing this up by any evidence.

Everyone who wants to talk about claude code raise a Jira ticket with steps to reproduce and please link to that.

Re: Claude Sonnet 4.5

#487

Post by ReverseCold »

simianwords wrote: post #210
pzo wrote: post #104

Earlier quoted context omitted.

it looks like its already there

It’s stupid… like just have a registry of models and let people automatically use them. It’s silly to wait for manual whitelisting each time for every app

It was there a few (<5? I think?) minutes after the Anthropic post went out. If you look at Windsurf's web traffic it looks like they did a thing (model is an int) to make it so the IDE doesn't need to update to get new models.

Re: Claude Sonnet 4.5

#488

Post by lxgr »

simonw wrote: post #377
rapfaria wrote: post #373

Earlier quoted context omitted.

His "pelican riding a bicycle" tests are now a classic and AI shops are benchmaxxing for it

They need to benchmaxxx a whole lot harder, the illustrations still all universally suck!

I fully expect a model to output a SVG made up of 1000x1000 rectangles (i.e. pixels) representing a raster image of a beautifully hand-drawn pelican riding a bicycle any day now :)

Re: Claude Sonnet 4.5

#489

Post by MichealCodes »

commakozzi wrote: post #290
SubiculumCode wrote: post #184

Earlier quoted context omitted.

I do wonder about this. I just don't know if it real or in our heads

It does feel like it has to be real. I've noticed it since chatGPT with GPT-3.5, once it hit big news publicly and demands were made on "censoring" its output to limit biases, etc. (not inherently a problem to do this with LLMs as a society, but it does affect the output for obvious reasons). Whatever workflow OpenAI and others have applied, seems to be post-release somehow? i'm ignorant and just speculating, but lit…

> I'm sure some of it could be in the parallelization of processing that has to occur to service the large amount of requests. and more and more traffic are spreading it thin?

Even if this is the case, benchmarks should be done at scale too if the models suffer from symptoms of scale. Otherwise the benchmarks are just a lie unless you have access to an unconstrained version of the model.

Re: Claude Sonnet 4.5

#490

Post by cloverich »

AppleBananaPie wrote: post #323
stavros wrote: post #289

Earlier quoted context omitted.

What was the irony? I thought you were referring to the "3x speed" part as the insane statement.

To me it seems like an arbitrary number that I'm not even sure how someone could accurately measure it but maybe I've missed something :)

It's actually not arbitrary! I measure my PR rate / ticket close rate before and after, which are generally tied to agreed on features / bugs (often user requested / reported ones). I think if it were commit rate or lines of code it would be less meaningful, but at least a (non refactoring) PR should indicate some level of increased user value / bug fix. Sure we could categorize it further and break it down more effectively, I'll not die on the sword of 3x, maybe its 1.5x, maybe its 4x. Neither seems a very meaningful difference when the comparison being discussed is 0x or even -X. The latter I think _most_ of the time is going to be prompt or task related, which is why I think its so important to share and discuss (particularly the negative case!)
Post reply on HN