HN user
Not the original commenter, but personally I noticed my quota usage didn’t feel like it was being spent at a much lower rate when using Sonnet even on a relatively low thinking budget and based on a few comments here it seems I might not be the only one. Has anyone else noticed this? Wasn’t it different in the past? I thought I would be getting to use Sonnet much much more than Opus but it did not feel that way despite being on 20x plan.
Do you have Usage credits turned on in your settings?
I was working with very small local models (<= 4B ones) in my app, there was a point where the response of the model seemed so good I really had to double check I wasn't mistakenly using a cloud model. The model hadn't made tool calls or anything, but its textual answer quality was high enough that it made me hesitate. A different kind of impressive vs the frontier models, but still was a bit of a wow moment when I remember expensive hosted models having a hard time writing proper English.
Billing caps are underrated! I don't understand why they aren't present everywhere. As an indie dev there are some services I'm really hesitant on trying by fear of getting an enormous bill for a mistake, this is even more true with vibe coding IMO.
I wonder why there aren't more open weights model with support for prompt caching on OpenRouter.
I see the Claude team wanted to make it less verbose, but that's actually something that bothered me since updating to Claude 4.7, what is the most recommended way to change it back to being as verbose as before? This is probably a matter of preference but I have a harder time with compact explanations and lists of points and that was originally one of the things I preferred with Claude.
That matches what I have seen, but I think I remember reading a tweet that had mentioned those "developing in the open" (not an exact citation, just based on what I remember), which made me wonder if it meant they considered this allowed only for open source software, or if they were intending to be much more permissive, essentially considering users can use their quotas wherever they want, or maybe even completely different rules, again I feel there could be more transparency regarding all of that.
Peter, while we are on the subject of clarifying what is and isn't allowed I have a question: has OpenAI clearly communicated about precisely where one is supposed to be able to use their Codex quota? For instance, as far as I understand, it is allowed to use it with OpenClaw, but does it extend to any other coding harness? Say I have an app (potentially a paid one) and want my users to use their Codex quota in it, is it permitted to do? As you can probably imagine that would unlock a lot of uses cases given smaller actors can't subsidize as much token costs, but unfortunately, and maybe expectedly due to the nature of subscriptions, I have not been able to find any answer regarding this.
I had a conversation right during the launch so not fully sure if it was Opus 4.7 but I also noticed the same behavior of asking questions that did not seem particularly useful to me, tho I still prefer that to not asking enough.
I'm not saying this should be every single domain. This isn't about products or management, instead I would frame it like this: I notice that multiple cases where we are worried about the impact of AI are basically just about the replacement of certain activities that some humans already aren't doing in today's society. If we are worried we will be less good at doing job X once we don't do job X anymore, why are we not worried about people who never did job X in the first place? If we are worried about people not doing jobs anymore, why are not worried for the human development of people wealthy enough not to work anymore for the rest of their days? I would not assume someone who won the lottery is going to have their life become uninteresting or see some cognitive decline. It could probably happen, but you can also see a path where the person just chooses to do the activities they always wanted to do, where they keep learning and exploring without the burden of usual life constraints. People already play chess when machines have beaten us for decades, just because they enjoy it.
Regarding education I think AI is a huge revolution waiting to happen. Usual courses have become boring? Have future super powerful AI generate per student highly personalized programs, create bespoke video games where succeeding can only happen once the student has validated all the notions you wanted them to validate etc.
Say someone uses AI, treating it as if it was a developer (probably not recommended today due to the risk of errors), and working and speaking with it as if they were some kind of product manager or senior engineer who only makes architectural decisions etc. I wonder what kind of difference would it really make? Sure the person might not be as good anymore as a developer, but how is this different from being a usual product manager or whatever the day AI truly is good enough for a developer role? I'm not saying I know what the answer to this question is, but this is something I genuinely wonder, and I think the same kind of questioning can apply to broader domains.
What is your app doing? Just LLM inference?
I hope some company trains their models so that expert switches are less often necessary just for these use cases.
Awesome! Are you planning on setting a license soon? I might have missed it but I don't see it on the GitHub repo.
Very interesting! On what platforms can this run? If it can run on iOS, how would you handle attempts to access to the file system or networking, is this already wired in somehow? If not is it easy to add custom handlers to handle these actions?
It is definitely not foolproof but IMHO, to some extent, it is easier to describe what you expect to see than to implement it so I don't find it unreasonable to think it might provide some advantages in terms of correctness.
I think both should be done, they don't really serve the same purpose.
It feel a bit like this to me. That's not to say LLMs should not have detected this, but I still feel like this fits the "vibes" the question gives, and some LLMs fall into that trap. Is it actually what's happening in the neural nets? Maybe not! But I always find it interesting or at least entertaining to approach those questions that way nonetheless; especially given the pattern matching nature of LLMs.
You are absolutely right! It's not just relevant, it's a much funnier take at robots mannerisms than what ended up having in the end.
The thing is that there is some overlap between trick questions and questions where the human is genuinely making a mistake themselves and where it would make sense for the model to step back and at least ask for clarification.
Oh no! Now it's going to be in the training dataset :'(
This is so elegant, especially with the art lights! To me, the desirable future for connected homes is one where technology is everywhere but mostly hidden and this is such a good example! This feels like an upgraded version of a chalkboard or sticky notes, but quite an optimal upgrade: one that makes it more useful and dynamic while keeping it mostly unintrusive.
On the subject of dedicated home control dashboards, I'm not sure I see their value at all given we all have screens in our pockets, so when it comes to enabling interactive controls I feel like using your existing devices or voice controls is the right approach.
I'm glad you asked because I must admit that in the last few weeks I totally thought this was just another agentic harness that happened to have a lot of extensions + ways to talk to it through messaging apps. So does this mean OpenClaw can connect to any agent? In that case I don't understand this part of the docs:
Legacy Claude, Codex, Gemini, and Opencode paths have been removed. Pi is the only coding agent path.
I wonder how many major applications and tools depend on sandbox-exec today despite that depreciation, IIRC I can think of the Codex CLI and Swift Package Manager.
Unfortunately I forgot which site it was and did not check if it was entirely paywalled, but if I find it again and it is I will let you know! Thank you!
I am a bit worried that this is the situation I am in with my (unpublished) commercial app right now: one of the major pain points I have is that while I have no doubt the app provides value in itself, I am worried about how many potential users will actually accept paying inference per token...
As an independent dev I also unfortunately don't have investors backing me to subsidize inference for my subscription plan.
The comment you are responding to is about ChatGPT/Codex, not Claude.
You are talking about Anthropic and indeed compared to OpenAI or GitHub Copilot they have seemed to be the ones with what I would personally describe as a more restrictive approach.
On the other hand OpenAI and GitHub Copilot have, as far as I know, explicitly allowed their users to connect to at least some third party tools and use their quotas from there, notably to OpenCode.
What is unclear to me is whether they are considering also allowing commercial apps to do that. For instance if I publish a subscription based app and my users pay for the app itself rather than for LLM inference, would that be allowed?
If you look at this tweet [1] and in particular responses under it, it still seems to me like some parts of it need additional clarification. For instance, I have seen some people interpret the tweet as meaning using the OAuth token is actually ok for personal experimentation with the Agent SDK, which can be seen as a slight contradiction with what you quoted. A parent tweet also mentioned the docs clean up causing some confusion.
None of this is legal advice, I'm just trying to understand what exactly is allowed or not.
Maybe I am missing something from the docs of your link, but I unfortunately don't think it actually states anything regarding allowing users to connect and use their Codex quota in third party apps.