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

Claude 3.7 Sonnet and Claude Code

anthropic.com

Post reply on HN
Print view View original

661–670 of 1001 posts

Re: Claude 3.7 Sonnet and Claude Code

#661

Post by allpratik »

babyshake wrote: post #190
bcherny wrote: post #91

Hi everyone! Boris from the Claude Code team here. @eschluntz, @catherinewu, @wolffiex, @bdr and I will be around for the next hour or so and we'll do our best to answer your questions about the product.

One thing I would love to have fixed - I type in a prompt, the model produces 90% or even 100% of the answer, and then shows an error that the system is at capacity and can't produce an answer. And then the response that has already been provided is removed! Please just make it where I can still have access to the response that has been provided, even if it is incomplete.

Plus one for this.

Re: Claude 3.7 Sonnet and Claude Code

#662

Post by SparkyMcUnicorn »

doctoboggan wrote: post #634
anotherpaulg wrote: post #331

Claude 3.7 Sonnet scored 60.4% on the aider polyglot leaderboard [0], WITHOUT USING THINKING. Tied for 3rd place with o3-mini-high. Sonnet 3.7 has the highest non-thinking score, taking that title from Sonnet 3.5. Aider 0.75.0 is out with support for 3.7 Sonnet [1]. Thinking support and thinking benchmark results coming soon. [0] https://aider.chat/docs/leaderboards/ [1] https://aider.chat/HISTORY.html#aider-v0750

Have you tried Claude 3.7 + Deepseek as the architect? Seeing as "DeepSeek R1 + claude-3-5-sonnet-20241022" is the second place option, "DeepSeek R1 + claude-3-7" would hopefully be the highest ranking choice so far?

It looks like Sonnet 3.7 (extended thinking) would be a better architect than R1.

I'll be trying out Sonnet 3.7 extended thinking + Sonnet 3.5 or Flash 2.0, which I assume would be at the top of the leaderboard.

Re: Claude 3.7 Sonnet and Claude Code

#663

Post by epistasis »

hectormalot wrote: post #557
epistasis wrote: post #457

Earlier quoted context omitted.

Update: Code tokens appear to be cheaper than 3.7 tokens, looks like it is around $0.75/million tokens for code, rather than the $3/million that the articles specifies for Claude 3.7

Likely because it is blended with cached token pricing, which is at $0.30/million. You can use ‘group by’ in the usage portal to see the breakdown.

Thanks, that's it. It's almost entirely "Prompt caching read."

Re: Claude 3.7 Sonnet and Claude Code

#664

Post by tw1984 »

pzo wrote: post #655
pclmulqdq wrote: post #604

Earlier quoted context omitted.

Also for $36.83 compared to o1's $186.50

But also for $36.83 compared to DeepSeek R1 + claude-3-5 it's $13.29 and for latter "Percent using correct edit format" is 100% vs 97.8% for 3.7. edit: would be interesting to see how combo DeepSeek R1 + claude-3-7 performs.

is there any public info on why such DeepSeek R1 + claude-3-5 combo worked better than using a single model?

Re: Claude 3.7 Sonnet and Claude Code

#665

Post by wokwokwok »

d_watt wrote: post #114

I'm about 50kloc into a project making a react native app / golang backend for recipes with grocery lists, collaborative editing, household sharing, so a complex data model and runtime. Purely from the experiment of "what's it like to build with AI, no lines of code directly written, just directing the AI." As I go through features, I'm comparing a matrix of Cursor, Cline, and Roo, with the various models. While I'm…

"no lines of code directly written, just directing the AI"

/skeptical face.

Without fail, every. single. person. I've met who says that, actually means "except for the code that I write", or "except for how I link the code it build together by hand".

If you are 50kloc in to a large complex project that you have literally written none of, and have, eg. used cursor to generate the code without any assistance... well, you should start a startup.

...because, that's what devin was supposed to be, and it was enormously and famously terrible at it.

So that would be either a) terribly exciting, or b) hyperbole.

Re: Claude 3.7 Sonnet and Claude Code

#667

Post by paradite »

mvdtnz wrote: post #479
k8sToGo wrote: post #368

Earlier quoted context omitted.

Have you tried just using the API and pay as you go?

That doesn't answer his very specific question.

It's actually fairly easy to setup a 3rd party app to use Claude via API, to get extremely generous limits.

I wrote a step-by-step guide for the app I built: https://prompt.16x.engineer/guide/claude

Re: Claude 3.7 Sonnet and Claude Code

#668

Post by SubiculumCode »

Squarex wrote: post #391
Workaccount2 wrote: post #386

Earlier quoted context omitted.

Having your own hardware to run LLMs will pay dividends. Despite getting off on the wrong foot, I still believe Google is best positioned to run away with the AI lead, solely because they are not beholden to Nvidia and not stuck with a 3rd party cloud provider. They are the only AI team that is top to bottom in-house.

I've used gemini for it's large context window before. It's a great model. But specifically in this benchmark it has always scored very low. So I wonder what has changed.

I don't know, but very recent Gemini models have certainly seemed much more impressive...and became my daily.

Re: Claude 3.7 Sonnet and Claude Code

#669

Post by silexia »

hubraumhugo wrote: post #131

You can get your HN profile analyzed by it and it's pretty funny :) https://hn-wrapped.kadoa.com/ I'm using this to test the humor of new models.

Roast You've posted so much about government waste that the IRS probably has a special folder just for your tax returns. Your hatred of VCs is so strong, I'm surprised you haven't built an app that automatically downvotes any HN post containing the phrase 'we're excited to announce our Series A'. You're the only person who reads the comments section on a post about electric vehicles and thinks 'This is the perfect place to explain fractional reserve banking!'

Re: Claude 3.7 Sonnet and Claude Code

#670

Post by zora_goron »

Does anyone know how this “user decides how much compute” is implemented architecturally? I assume it’s the same underlying model, so what factor pushes the model to for longer or shorter? Just a prompt-time modification or something else?
Post reply on HN