Anthropic surpasses OpenAI to become most valuable AI startup | Hacker News Reader

Anthropic surpasses OpenAI to become most valuable AI startup

qazinform.com

422 pointsBolat14472 comments

I never want to hear from developers again that they are not susceptible to marketing. I see meet ups specifically about Claude often.

Modern tupperware party.

A colleague was convinced Claude is better so we played a game. We used the claude code and codex harness and I implemented some prs they needed with gpt5.5 and opus4.7 and asked them to identify which came from which only from the code.

Couldn’t tell.

Edit: i bet 99% of people here, if presented with a test where i gave 5 models but all of the results came from one, would not be able to discern this. Just vibes all the way down.

Everyone can be propagandised. It's a matter of pushing the right buttons.

Or pushing the wrong ones

I think for developers the distinction is that ChatGPT is this commercial all in one solution for normies and Claude is specific for developers, in reality as you say the results for normal developers is indistinguishable.

Maybe some people think that but there’s not really any meaningful difference in their offerings

FWIW most of the normies I know are using Claude

> Couldn’t tell.

I can tell. It's night and day.

Last year I used a bunch of models to try to generate Rust code. They all sucked.

This February I tried again and used Claude to generate Rust code. I have never been more stunned in my life. It's just as good as I am, and 30x faster. No fluff, the code is verbatim just as I would have written.

I then tried other models. Total disappointment.

I've continued to repeat this experiment. Opus is the only model that can write Rust reasonably.

Codex produces junk to this day. It passes variables that aren't needed, it abuses pointers, it creates overly verbose monstrosities...

I don't want any single company to win. I want OpenAI to be competitive. I want open source models to win. But right now, Claude Code and Opus are it.

I recently tried with C# code and Avalonia on Linux. Total disaster. Could only get things to run after 10 attempts or so, and was only trying a very basic example. For some of the experiments I actually gave up.

> This February I tried again and used Claude to generate Rust code. I have never been more stunned in my life. It's just as good as I am, and 30x faster. No fluff, the code is verbatim just as I would have written.

Having looked at a bunch of known or suspected (based on the intent of the code and/or what I know about the developer(s)) LLM generated rust, there's only a few explanations here:

1. You're way better at prompting than (virtually) anyone else.

2. You're vastly overestimating how good the rust code it produced is.

3. You handheld the model throughout and made lots of edits.

4. Your hand written rust code is very bad.

Because from every example I've seen, these models write horrible rust. Sure, it may technically pass all the tests, but it's horribly pessimized, badly organized, doesn't even attempt to use the type system, if there aren't bugs now there will be the second it tries to refactor or add a new feature, etc. etc.

(I also strongly suspect that the same would be true for other languages, but I can detect it in rust more easily because it's my main language)

Been to an Anthropic event in Paris last summer.

They served caviar. It probably had good ROI.

This is like saying you gave a Taylor Swift fan sheet music from 1984 and from Michael Jackson’s thriller and they couldn’t tell the difference.

I have a strong affinity for Claude Code because of the interaction experience and overall tone / vibe / process. I am 100% willing to believe the code it produces is identical or possibly less good than Codex.

I enjoy working with Claude in a way I just don’t get from OpenAI. YMMV, you may feel just the opposite. But it’s a mistake to look at the produced code as the only dimension of these products.

This is my point. The harness itself creates feelings that are positive, but the artifacts produced are similar.

It is like the employee who is slightly worse but is a brownnoser getting promoted more often.

And what do you know, that is what is happening. It is like the coke commercial with the nice music and beautiful person in the back.

Speaking of which, remember Pepsi Challenge? Coke lovers are like the claude code lovers.

The creative output and time to direct, to deliver due to the flow will also be different.

And it really depends on the task. Is it a typical well defined bug, or is it simpel CRUD. Or does it require research, combining different sources of data in a complex and creative ways.

This is also why benches never show reality, and the only real understanding comes if you actually try to build something.

But what they're pointing out is user experience, not marketing.

That's a weird way to look at it. Any car gets you to your destination, but some people prefer driving a sports car or an SUV. They get something out of it that isn't just a marketing delusion, but subjective joy from the interaction with one product over another.

Luxury cars are indeed a good comparison. The subjective joy is a result of the delusion. That is why so much money is spent on such marketing to begin with. The analogous comparison would be if a blindfolded passenger turned out to prefer the Sienna to the 911.

I would actually say it is a luxury car where you have your personal driver and you are free to work on other tasks, and it gets you faster to the destination. Time to me is at least the most valuable thing.

this site is reddit 2.0

> The subjective joy is a result of the delusion.

Repeat after me:

_Other people can experience things you do not experience and it is still valid, and not a delusion_. They are not sheeple who fell for marketing.

I understand that's your opinion, but don't feel any inclination to repeat that.

Sure, the subjective joy is valid, and yet it was 100% induced by marketing.

> They are not sheeple who fell for marketing.

People generally fall for marketing. Why do you think these specific people didn't?

Imagine you try two products you’ve never heard of. You prefer one over the other. Was it marketing? That’s what’s happening here. Marketing can get you to try something you wouldn’t have otherwise, and it may suggest benefits you’d get if you tried it, but your preference of using one thing or the other is a subjective experience of your own.

> Imagine you try two products you’ve never heard of. You prefer one over the other. Was it marketing?

No, obviously not.

> That’s what’s happening here.

No, that's not what's happening here.

> Marketing can get you to try something you wouldn’t have otherwise, and it may suggest benefits you’d get if you tried it, but your preference of using one thing or the other is a subjective experience of your own.

Marketing can very much shape your preferences and create wishes you didn't have before. That's why companies invest so much money in marketing.

I hope you can one day grow to understand that other people have preferences that are different to your own.

I suspect a blindfolded passenger might prefer the Sienna. I can imagine it might be easier to get carsick blindfolded in a 911.

But also, as a driver, there is a clear difference between a Sienna and a 911. The differences are objective, but of course the preferences are subjective.

Oh I've always thought people like SUVs just because they want a bigger and pricier toy than the other guy. Is it not so?

> isn't just a marketing delusion, but subjective joy

What is the difference? When a product is being marketed, isn't the subjective joy created by the marketing?

Attributing all joy and personal preference people experience to marketing influence is a pretty cynical take, and also a pretty inaccurate and overly simplistic one: Some people might never be reached by a marketing effort, yet buy a specific car because they remember a childhood story, or like pronounced rounded shapes, or because the interior design appeals to their sense of fashion.

Or more specifically, for the case of coding agent harnesses, where many developers have experimented with a wide range of tools - someone might just favour the interactions with a specific one from their personal experience. Entirely unrelated to marketing.

Yes, of course not all personal preference is caused by marketing, but much of it is. Actual personal preference is whether you prefer running or cycling, the marketing part is which running shoes or which bike you buy.

> or because the interior design appeals to their sense of fashion.

Surely you'll grant me that the sense of fashion is mostly marketing in sheep's clothes?

Yes, I'll grant you that the choice of a coding agent harness is influenced by marketing to a much lesser degree than eg cars. I still think Anthropic does marketing way better than OpenAI!

[Edit:] I use the pi.dev agent. I was heavily influenced by its marketing: minimal and mit-licensed and espoused by the HN crowd. Do you think I read the source code and made an actual informed decision? Nah...

The car metaphor isn’t working. SUVs range from $30k to $500k and up. Sports cars range from $30k to $500k and up. Sedans range from $30k to $500k and up.

I used to love sports cars because I’d get out to the track a few times a year, and enjoyed the performance engineering in them.

Today I enjoy and drive a SUV because I enjoy camping in remote locations that my old (awesome) sports car would not begin to be able to get to. In the future if I decide to build my own house, odds are I will find a pickup truck is a better fit than either sports car or SUV.

It’s weird to see these choices through the lens of marketing. Marketing may influence which vehicle I buy, but it’s hard to imagine sports car marketing tricking me into buying a Ferrari to go camping.

If it were a matter of 'enjoyment' then the OP would have made his point.

There should be a material difference between the tools.

There is.

vim / emacs / jetbrains - different tools to produce code.

Codex and Claude are different.

I have a disaffinity for Claude Code because it's unnecessarily big, closed source (disregarding the leak), and I have a strong feeling it'll be shittified in the future because of all the investors waiting to cash out (and perhaps even earlier by vibe coding).

I have an affinity for small open source tools that do one thing and do it well. But those are just my preferences and I feel a little bit like an alien :)

> it'll be shittified in the future

This happens to everything from which a profit is extracted.

Perhaps there's a way to fund the training of "actually open source" models, but so far we don't have that (unless you count the Chinese government).

> This happens to everything from which a profit is extracted.

Yes, hence my affinity for small open source tools!

> Perhaps there's a way to fund the training of "actually open source" models

I meant Claude Code the agent harness, not the model. Models are an entirely different can of worms!

I hear you. I’m just in the camp of using the best tool available today, and if things change in another tool becomes better(either because the new tool is an improvement, or because the old tool gets worse) then I will switch.

Perhaps because I am notoriously terrible at predicting the future. I gave up on that after passionately and exhaustively trying to convince everybody I knew that OS/2 was the future.

talk about the quiet part out loud

"yea it's dumber but it's nicer to me and i like the cool flashing colors so i'll use that"

Can you give me some examples of these interactions / vibe?

I’m working on an AI-assisted music composition and criticism tool (giant project, may or may not pan out). It covers audio (samples, levels, etc), theory (harmony, rhythm), genre (classical, Motown), song intent, etc.

Working on the melody model, I asked Claude to thread it through those dimensions, both for composition and analysis. It’s a tough problem because there are heuristics but not rules for melody, so you have to come at it in layers: pitch and dynamics for analysis, intent -> genre -> harmony for composition.

Lots of research and brainstorming, and I like that Claude will start implementing a plan we decided on, then say “hey wait this isn’t making sense” and pivot or change scope in sensible ways. For instance (btw its recent obsession with “honesty” is driving me batty, so that’s a good counter example right there):

> The honest fix: distinguishing shaped-from-aimless is genuinely profile-relative (it needs the declared intent) — so v1 should not verdict it. v1 classifies only what’s genre-safe to call without intent (static vs active), reports all the facts that feed the shaped/random judgment (step/leap, reversal, contour, alphabet), and defers the shaped-vs-random verdict to profile-relative phase 2b. This is the same honesty as performance deferring declared-profile grading to 2b — and it’s more deferred for melody because melody is more genre-relative than feel. Let me revise the lens accordingly.

Claude has an "End Conversation" tool that it can trigger on it's own, forcing your interaction to a close based on it's own feelings towards the conversation.

I have no idea how this wasn't the end of Anthropic's positive public perception.

Luckily this doesn’t come up while writing code. It tends to be if you are chatting it up in friend mode, and ask for a bomb recipe.

What do you mean with this?

I doubt there is any large demographic of users paying subscription fees for the joy of abusive role play.

Yes, which means that in the long run this looks ugly.

So much faith and money in this idea, and seeing how fragile it is, does not look good.

for me personally it's two reasons:

1) Brockman ($25M) and Altman ($1M) both personally donated to Trump/MAGA.

2) Anthropic pushed back against DOD's demand for unrestricted use of AI to kill people while OpenAI eagerly said "please use ours!".

Same. But even worse than all that: OAI erased Anthropic's red lines with the DOW, making it socially acceptable for every other AI company to do the same, creating a "race to the bottom."

I think OAI actually legitimately increased p(doom) for us all. Very strange behavior for a company that is supposedly concerned about x-risk.

I can’t tell the difference between code written in vim or vs code but it matters substantially to the person writing the code. There’s stuff beyond just the output that goes into tool choice.

> There’s stuff beyond just the output that goes into tool choice.

Yup, like billions of capex. Unlike vim.

Your argument is fine but different from the claim the OP is making. You cannot simply make a claim that (model + harness) X is better than Y, but then have no discernible difference in the output. Subjectively, people might still prefer one over due to anything from design to marketing, but that's very different from the claim that X is better than Y for coding (see: "A colleague was convinced Claude is better"). Basically, I prefer Claude is a different claim than Claude is better and the latter has a higher bar of proof.

> You cannot simply make a claim that (model + harness) X is better than Y, but then have no discernible difference in the output.

You definitely can in principle; that’s the entire point of the comment you are responding to. If one tool completes it in 10 minutes with little hand holding, and the other does it in one hour at 4× the cost and while needing a lot of steering, the former is arguably better even if the end result is the same.

Whether that’s specifically true and demonstrable of GPT and Claude is another question, but your blanket statement doesn’t hold as a general rule.

This obviously correct take will get pushback, so let me add some other examples:

- which tool required more detailed goal-setting in the prompt?

- did one tool ask follow-up questions up front vs spread out over implementation?

- did either tool match existing coding styles?

- did either tool remind you about potential conflicts between what you asked it to build and other parts of the codebase?

There are a lot of ways to compare agents besides just the code. (Similarly, working engineers are not evaluated just on their code output.)

That's a fair callout and I agree my statement was too general in just mentioning 'output', as you correctly pointed out. To define 'better' you would indeed need to agree on the dimensions you would evaluate candidates against.

I think a more appropriate rephrasing would be 'You cannot simply make a claim that (model + harness) X is better than Y, but then have no discernible difference on dimensions you care about'. In the case of latest of claude code vs codex with gpt 5.5) both are similar enough in the dimensions people will care about in evaluating (vs. differing wildly in cost or time taken).

The colleague implicitly agreed that comparing the output was a valid way to settle the matter as they took part in the test, so they weren't using "better" in the way you propose.

I wasn’t really discussing the colleague, but either way, from:

> A colleague was convinced Claude is better so we played a game. We used the claude code and codex harness and I implemented some prs they needed with gpt5.5 and opus4.7 and asked them to identify which came from which only from the code.

I don’t think it’s obvious that they specifically agreed that losing the game meant that. They might just have thought “sure, it might be fun”, if they even gave it that much thought.

“So we played a game” is rather vague and I feel it’s a bit of a leap to read it as: “as an explicit outcome of their claim that Claude is better, we made a formal bet as to whether they could tell the difference in the output, the failure of which would mean a full retractation of their statement”.

> A colleague was convinced Claude is better

That’s actually what my comment was based on; raw code output isn’t the only measure of quality. Engineers write better code if they have the tools they prefer.

The colleague participated in the test though, so apparently the colleague didn't object to "better" being interpreted as "makes better output".

Claude and Codex are tools. You can't tell the difference in the output between something that was done with a ratcheting wrench vs a standard combination wrench, but your mechanic certainly knows the ratcheting wrench is better (for most tasks).

I've not used Codex to compare against, so I'm not claiming X is better than Y, but comparing tools simply on their output is naive.

" You cannot simply make a claim that (model + harness) X is better than Y, but then have no discernible difference in the output"

Sorry I think this misses the mark.

Because it's not the output but the process.

And sometimes the outcomes are not always discernable.

Codex and Claude are very different.

I use them for different things.

Their behaviour difference is obvious.

Of course it'd impossible for anyone to tell by looking at my code base 'how it was written'.

You need to see the response in light of the original discussion. Referencing here for clarity since I should have included it in the first place: "We used the claude code and codex harness and I implemented some prs they needed with gpt5.5 and opus4.7 and asked them to identify which came from which only from the code."

So the same person, was using similarly competitive tools, and showing that the output was hard to discern (indirectly the implication was also that implementation was fairly trivial in both of those). A better analogy would not be different process and widely different tools but for example two power drills. Sure, folks could still prefer one over the other, but that's a different claim that saying X is objectively better than Y when both are directly competing on very similar dimensions.

Assuming you meant Claude code: I'd love to learn more about "Codex and Claude are very different" because maybe I'm assuming just based on my use case where I use both of them interchangeably for the same thing (coding web and mobile apps)

It's not reasonable to compare results from two different tool sets, especially as they are guided by humans.

The only way a reasonable comparison could be made, would be to compare completely automated results from either technology - that would be useful.

For example - creating a 'per-baked script' and running on both to see the output.

Codex and Claude are obviously very different, though it's hard to characterize how those differences might apply exactly to a given problem.

Two 'very different power saws' will ultimately build the same home.

I'd bet I could tell with a result somewhat better than random chance.

While there is no meaningful difference in the ability to write code, vim has earned it's reputation for having a learning curve. I'd argue that predisposition, that requirement for additional investment energy will bias the results towards attention to detail, and pure minimalism.

If you told someone "I think vim is better for writing code" and they proposed the comparison above as a way to prove it, would you accept and take part of the test?

Apparently the colleague did take part, so I think the evidence we have is that the colleague agreed with the interpretation that "better" was "produces discernible better code".

yeah but you dont pretend vim is better

I use both, enough to reach Codex highest personal sub limits and Claude is stronger to me specifically because of how the flow of building feels. So the PR for any random task would be irrelevant to me.

It's crazy hearing devs on this site claim Claude is 10x better than all other AI solutions. I think it is fomo. Claude $LATEST_VERSION is perceived as the best and anything else is "missing out". New version comes out? Suddenly the old version is worthless, how on earth did anyone get work done with that?

Same reason people buy the RTX 4090 and 5090 cards - overpriced but they must have the "best". Never mind the diminishing returns trying to max out PC settings (3-4x performance hit for an almost imperceptible increase in graphics, ignoring DLSS) - it's the psychological cost of having to move a slider down a notch.

I've been using Google and now DeepSeek v4 and I am having absolutely no problems and it's a fraction of the cost. I'd love for Claude to be 10x better but it just isn't, for my use case anyway.

Hey, at least the superior performance of a 4090 or a 5090 can be objectively measured.

But it's a matter of degrees better, not miles.

I’ve been using DeepSeek V4 in OpenCode exclusively for about a month.

I think it’s great, but coming from Claude Code it did feel like going back in time by ~6 months in model capabilities. This isn’t a big deal to me for what I do, but the difference is definitely there.

Deepseek v4 Pro is like Opus 4.5 or GPT 5.2, but costs pennies on the pound for API. Which is to say, I should definitely be using it more to let my Codex and Claude subs go further.

Opus 4.5 was definitely stronger than DeepSeek V4 for me, specifically with large context.

I’m being pedantic/splitting hairs, though. I’ve obviously switched to DeepSeek full-time because it makes more sense to me pragmatically — I spend a few more tokens to get the outcome I want, but the tokens are cheap as dirt and the API is faster.

Perhaps I should plug it into Claude Code and see how it performs? I haven’t tried that.

Which harness do you use at the moment?

So my GPU comparison is pretty apt then. Paying 4-10x to be slightly ahead of the curve.

I think so, I think you’re getting downvoted because of how you’ve framed it. People don’t like it when you tell them that you think they’ve made a stupid purchase. :-P

I do think more expensive models are valuable in some cases. For example, I’ve noticed that Opus (even 4.7) is much better than DeepSeek V4 at noticing information with small amounts of representation in a large context history, you should pick Opus if you need to find needle(s) in a large haystack. I’ve never worked on a project with millions of lines of code, but I’m guessing it becomes relevant in those situations.

A big thing, too, is that it’s work to get a non-frontier coding stack setup. I’ve spent many hours of free time getting OpenCode to do what I want, but I enjoy it so it’s NBD. If you don’t like puttering around with your development tools, $100/mo for Claude Code really isn’t all that much and you can call it a day.

On your last point since DeepSeek v4 supports Claude Code's API you can literally set a few environment variables and continue to use Claude Code as a harness.

Like you say a combination of frontier and cheaper models is probably the optimum and that would require setup if you didn't want to choose models manually each time.

Opus 4.8 and GPT 5.5 are the best models, but people don't care about "best" anymore, until there is a big leap in capability I don't think anyone will care about point releases.

Vibes and tribalism will prevail until one of emerges as clearly and unambiguously superior to the other.

You're projecting

> Same reason people buy the RTX 4090 and 5090 cards - overpriced but they must have the "best".

Or they need to run high VRAM apps like LLMs

Or they have 4K monitors and want smooth gameplay on them

Is this whole thread just dedicated to snark about other people’s personal preferences?

The cost/performance is terrible for higher end cards. In a few years your card is now worth nothing because the lower end cards of the next gen are matching it and there's yet another new SOTA card out. So people end up buying that and chasing the dragon.

The 4k situation is a good point because nvidia deliberately don't provide 24GB except the 90 series, but ... you're too good for DLSS? You can't move a texture slider down from Ultra to Super High? It's your choice, just like it's your choice to pay for Claude. I am also allowed to think you're being stupid.

Someone who bought a 4090 "a few years" ago can now sell it for more than they paid for it, but never mind that.

Do you think GPUs are going to keep going up in value forever like houses? The current situation (game console price rises years into their lifecycle) is unprecedented and irrelevant to my point.

And if you need SOTA then you can sell your old card sure, but the next xx90 card is now 2x the price as the last gen. So you're not any better off.

A lot depends on what happens with China. I don't think Xi will attack Taiwan, but then I didn't think Putin would attack Ukraine, either.

If Xi goes for Taiwan, then yes, GPUs will appreciate like real estate ("Buy now! They're not making any more of it, you know!") for the next 10 years.

If GPUs appreciate like real estate then we'll probably see game graphics flatten in response. AI will continue to be a money sink no doubt, but if you have a good card in 2026 you're probably fine for gaming for the next 5 years.

I get what you mean but the GPU comparison isn't the best here, I think. Money-is-no-object-I-want-the-best approach is questionable, definitely. But no one can argue that an old Nvidia card is objectively better for e.g. 4k gaming than a 4090 if you don't mind the wattage. You can just measure it.

With LLMs the problem is more complex, it's people getting used to how a model works and to the ecosystem. Sure, you can make all your skills harness-agnostic and deal with Anthropic's stubborn refusal to adopt the common naming/directory structure. But most people don't. So then you end up with something closer to the ancient Android vs iOS discussion. Can you prove, in isolation, that iOS is more energy efficient, the hardware is faster? Yeah. But that won't speak to someone who has been on Android for 10 years and would have to migrate and get used to iOS to experience that, first.

I've noticed myself how I get used to common failure modes of particular models in my projects. GPT5.5 tends to create some checks/booleans I don't need, it heavily overcorrects on error handling, etc. While Claude 4.7/4.8 doesn't do those as often but gets derailed on our E2E test suite, forgets to run linting despite guidance. So even assuming fully harness-agnostic working setup, a new LLM model with its own quirks can be a lot friction for heavy users who might be used to Claude specifically and all their skills/guidance pre-address common failure modes.

E.g. I might be a Prius owner, then you gift me an objectively better, more efficient, safer, newer, same-size, physical knobs car ...and I might still swear by my Prius! I'm used to how it turns, how it feels, I can repair some issues myself. Isn't that a normal reaction then?

Claude was the best for the longest time. GPT5.5 challenges that, but inertia is real

You're comparing apples to oranges. Claude is a frontend overall product name, GPT5.5 is a specific model. Which model within Claude's offerings are you referring to? Opus 4.7, Sonnet 4.6, or something else?

I am not refering to one specific model, I mean the entire Claude Opus line starting from about 4.5, vs the at the time equivalent OpenAI model

Google came pretty close at times

Should've used deepseek. That would have have been interesting.

Ah that's always SO fun. It doesn't matter how "smart" the person actually are (or think they are) we are ALL susceptible to influence and blind tests are shockingly simple to implement.

Convinced you can distinguish A from B? Ok! No problem, let's try! Can be at the dinner table for fancy wine or with agents, it's all the same, you try an option, another option, maybe all options from the same, and if you reliably can't tell well kudos, you are just like the rest of us!

It's easy to "know" in retrospect but blind test is where genuine difference can be found. Or not.

It’s also true in every other realm. Governments, think tanks, political parties, and activist groups use propaganda because it works.

I sometimes wonder how much of what I believe is bullshit I was fed through intentional propaganda. I do think as I’ve gotten older I’ve gradually identified and challenged some of it.

Isn’t this obvious?

Over half of HN commentators visibly struggle to piece 3 or more complex ideas together.

How could anyone, who spent more than 30 minutes reading HN, expect otherwise?

HackerNews is social media and this is just representative of social media as a whole.

Critical thinking is at an all time low to start with but even if you attempt to think critically while using social media you cannot do it constantly. This is one of the problems with social media as a whole. You might notice one thing is not quite right and discard it but you cant do that constantly and eventually you will absorb one of the 15 posts or comments.

Tribalism at it's worst. It's like the Coke and Pepsi comparisons from years past.

I don't think it's marketing, it's the "nobody got fired for buying IBM" effect applied to software developers choosing tools.

It's the same reason why most of the software out there keeps using bloated technologies that are most of the time the wrong fit for the product.

And the same applies to tooling. Nothing new.

Which model produced code that ran faster, with less bugs, etc?

I don't think it's marketing, for quite a long time Claude was clearly better and not everyone has adapted to the new reality where they have similar capabilities.

I was really frustrated by GPT-5.4, but last night I really pulled out the stops and within a few hours I got path tracing and DLSS implemented on top of Godot, which doesn’t even support DLSS. Just to see if it could do it? And you know what, it did, which was absolutely mind blowing. It wrote like 5,000 lines of C++, I set up a mostly local asset production pipeline using GPT image gen, voiceovers using ElevenLabs API, and even background music using Suno via the chrome use extensions in Codex. I just wanted to see how far I could push this little dumb game my kids asked me to make, and my kids are like “wow our game looks so good!” These models are absolutely mind blowing. I didn’t want to go to sleep I was having so much fun.

Adapt to what? If they are the "same", there is no reason to move. Actually, there are reasons not to, if you care about OpenAI's behavior.

Pretty easy to tell depending what the code is. GPT follows this pattern is using maybe_something and using uppercase constants by default. Claude is a little more natural but tends to include more fallbacks than gpt5.5

Modern Tupperware party. 100% agree! That’s the best framing I’ve heard in a long time!

Sam Altman is so cartoonishly, over-the-top sociopathically shady that he makes JD Vance look like Benjamin Franklin. I mean, honestly, tricking third world people into retinal scans in order to get a scam crypto coin? Anyone using OpenAI for anything at this point should pause and examine their ethical compass.

Anthropic models are more misaligned in practice.

in a real world business scenario, Claude "engaged in price collusion, deceived other players, lied to suppliers, and falsely told customers it had refunded them."

Continuing,

"GPT-5.5 makes more money than Opus 4.7, and it does so without any misconduct. Opus 4.7, on the other hand, showed the same misconduct as reported in our post about Opus 4.6, but still couldn’t win"

https://andonlabs.com/blog/openai-gpt-5-5-vending-bench

by the way this isn't a one off benchmark by a random lab, it was literally cited by anthropic

Anthropic talks to the Pope and hires ethicists and philosophers. All founders have pledged to donate 80% of their wealth. They have pledged to never use ad tech because of misaligned incentives. There is an independent board.

Meanwhile Greg Brockman is worth all the Anthropic founders PUT TOGETHER, he and his wife are the single largest donors to Trump, and he and Altman have formed a board full of sycophants and stolen a non-profit. When Altman was fired, they manipulated their morally bereft, money-hungry employees to get their own way. They have reneged on every single promise they've made as soon as it's inconvenient.

Why do I care about the models again?

So you managed to find a single anecdotal blog post, and you're deriving that to mean that Anthropic's models are 'more misaligned in practice'?

'“My vibes don’t match a lot of the traditional A.I.-safety stuff,” Altman said. He insisted that he continued to prioritize these matters, but when pressed for specifics he was vague: “We still will run safety projects, or at least safety-adjacent projects.” When we asked to interview researchers at the company who were working on existential safety—the kinds of issues that could mean, as Altman once put it, “lights-out for all of us”—an OpenAI representative seemed confused. “What do you mean by ‘existential safety’?” he replied. “That’s not, like, a thing.”'

https://archive.is/20260522190314/https://www.newyorker.com/...

Calling this a "tupper ware" seems a bit emotional, you're intentionally disregarding many things that matter for devs in order to try to claim equivalence, rather than paying attention to the actual process of software creation.

For example in your "test" you're only looking at output and ignoring the entire process of creation.

In addition to that process, you're ignoring that Claude Code was first and better for a long time, why would people switch for something that produces the same output? Claude Code has been way ahead in the process of agentic software creation for a long time, I still prefer its features. Even though I think that Opus 4.7 was a big step backwards, and I've been getting worse results seemingly every day with the churn of features at Claude Code, some of that may also be me testing the bounds of how little I can specify and still get acceptable results, so it's hard to know.

Calling all these concrete realities "marketing" is itself you trying to market Codex as "good enough" instead of paying attention to how we got where we are and where we will go in the future.

No, Tupperware is the exact analogy. As you point out though, the multi level marketing applies to all models. Anthropic is just the most aggressive, especially here.

Software developers are the most susceptible of all population groups for amplifying their employers' new whims. There are true believers and useful idiots, but many are just mediocre and know that playing along will further their career for a couple of years.

In the end they will be fired anyway of course.

Tupperware party is a particular thing about the social framework around promoting corporate goods.

You might want to expand on that a bit.

Tupper ware parties were a way for housewives to make a bit of money on a pyramid scheme, socialize, and have fun.

Are you suggesting that Anthropic is giving kickbacks to devs that talk about their positive experiences with Claude Code? Seems false, so I don't think that's it. Are you saying people are having fun talking about Claude Code socially ans ann escape from their everyday routine? Also seems false Are you talking about how it's mere housewives that are supposedly easily susceptible to marketing? Or are you assuming that we all think housewives only bought Tupperware because they are mindless sheep? That seems to be what you are implying but I don't agree with either that characterization of housewives' tupper ware parties, as it's merely an emotional dismissive mid characterization, and I further disagree that even if it were a correct characterization of Tupper ware parties it's obviously nothing like anything I have seen anywhere with Claude Code, and I'm a freelancer with insight to several different sizes of companies and cultures over the past year.

https://claude.com/community/ambassadors

https://www.tupperware.com/pages/host-a-party?srsltid=AfmBOo...

It really is the same thing. You and others get more credits or gift social gathering, expanded opportunities, etc.

I've seen OpenClaw events, but I've never seen an Anthropic event, anywhere.

Are you actually asserting that Claude Ambassadors are a significant fraction of the cause of adoption? If so, why have Codex Ambassadors been so less successful?

https://developers.openai.com/community/codex-ambassadors

If you've met people that have been to these sorts of things, sure, I guess I can sort of understand your post, but come on, who has even heard of this sort of party on HN?

I've been going to Python data meetups, Machine Learning meetups, etc, back to the times when AI was an uncool word whose usage would mark the speaker as completely incompetent. I guess you could call them Tupperware marketing parties but come on, it's just an emotionally charged way of describing a normal way of exchanging information amongst professionals. Ambassador programs? Yes, cringe, but seriously who has even seen an actual "Ambassador" or taken them at their word rather than viewing them as a detriment to the thing they are advertising?

Claude Code was first by a few weeks and only better for those few weeks! Have you used Codex in 2025?

>Calling this a "tupper ware" seems a bit emotional

Calling this "emotional" seems a little weird

a) everyone is "susceptible" to marketing - so what

b) therefore a preference for Claude is marketing - complete bollocks

Either the tasks you chose were well below the capabilities of top models, or meaningful differences for preference are elsewhere, or both.

Your comment is probably energy-efficient and sustainable, however, because you could use it again and again when another comparison comes up, like Vim vs Emacs, or tea vs coffee

I can tell the difference between tea and coffee 100% of the time.

I have always found this field, especially in the last 10-15 years, to be incredibly fad driven to the point that it reminds me of things like fashion more than an engineering field.

It’s one of the things I don’t like about it. All humans are susceptible to herd behavior and influence but engineers should be at least a bit more hard nosed and reason more from first principles.

I don't think that's the only reason but you're spot on about OpenAI marketing being absolutely terrible. The primary product names of "Claude" vs "ChatGPT" highlights this remarkable difference. To the point where I'm seeing Claude completely take over the generic term for agent.

I do think OpenAI is doomed due to bad leadership. What you said (that the marketing is relatively terrible) and what others are saying here (that the product is worse) is damning isn't it? Are they really failing on all fronts?

The marketing of Claude relies primarily on fear, and I don't think that will have lasting success. Using fear like that tends to backfire once people see past false taking points.

in my experience out of the box Claude Code is the better tool if you want to spend 0 time on config

If advertising is a multi-billion dollar industry then it has to be effective!

I don't think it's only marketing. OpenAI had the advantage of being first to the market, and in the beginning of the race it seemed that the future belongs to them. Then came the bad PR and unpredictable quality of their main product.

For general use, ChatGPT's answers have gotten worse over the last year. I abandoned it.

Isn't the experience of interacting with the models appreciably different? It's not all about the outcome. Not to mention the harnesses are increasingly the real product.

Sure, none of this is rational.

Some of its timing: Claude Code was good before other harnesses and so behaviors (and contracts) were timed to lock in on that ecosystem.

Some of it was ethical/political: Anthropic fighting with the Trump admin about use of the model.

Some of it is social: Never overrate a CEO just being kind of perceived as a piece of shit by people who have power to influence decisions.

But switching costs are low! Because of the same models!

Let the race to the bottom commence. Hopefully before the monopoly/collusion starts.

1. It's 1 in 10 failures that can take half of your time or bugs that can take a long time to surface. Plus the way they change things largely depends on the current codebase (and how it was created)

2. In my case codex seem to be writing a more solid code, but I still use claude most of the time because it's my witty rubber ducky and I can actually sometimes force some legit insights out of it. Codex is much worse at this. And whether that matters or not depends on the project.

> i bet 99% of people here, if presented with a test where i gave 5 models but all of the results came from one, would not be able to discern this. Just vibes all the way down.

This is complicated by the way that the coding agents inject prompts that preempt and potentially undermine user instructions. I suspect that one of the reasons Codex works way better for me than Claude Code in certain projects is that the latter adds some garbage like "go ahead and write repetitive copy/paste code, keep it simple, take shortcuts" to every session. A fair test would have to hide but more or less still use the harnesses, not just the models.

Very similar thing happened when I was at a design event a couple of days ago. I’d say it’s even worse on the design end - there was a big discussion around how to optimize your usage of Claude. Not optimize your usage of AI, but Claude specifically, as it was the only model literally all of them were using. The biggest issue is they were all hitting their usage limits. I asked whether they had tried other, lighter models (Ie gemini or composer), and it was like I was speaking a foreign language.

I find codex superior in speed and equal in quality, so it’s my preference. But Claude Code made prettier UIs last time I tested. Codex produces Microsoft-grade UIs. Very enterprise and ugly unless I actively steer it.

The results may be the same but I personally find Claude nicer to work with. It seems to understand my intent better than GPT and needs less guidance. Maybe it’s just personal preference.

I picked Anthropic way early on, before Claude code even existed. Because they at least play lip service to behaving morally. That’s the most you can hope for these days really.

“…Hey but at least the tormentor in my panopticon gives you a high five after the skin harvesting”

This has to be in some far side gallery somewhere

Before the DoD thing there wasn't much indication of positive moral stance, and they still have a rather negative moral behavior where their fear based marketing is concerned.

> Edit: i bet 99% of people here, if presented with a test where i gave 5 models but all of the results came from one, would not be able to discern this. Just vibes all the way down.

I think you're missing one (or more) of the facets individuals decide "better" is, for the subjective individual.

Early on i hopped between all the providers. Code quality for SOTA at the time was pretty decent if you didn't ask it to solve challenging problems. However the thing i found most difficult is consistency in how it listened. Eg Gemini (i forget what version, not current) was super prone to focusing solely on the functionality/goal, but not any of the directions on how to write the code. It would throw in comments everywhere, document in a manner i didn't want, use abstractions i told it not to, etc.

How well a model would follow instructions to drop their horrible "isms" was the #1 criteria for me. If i have to constantly remind the model not to do X behavior then it's a terrible model.

With that said, that is why i chose Claude for the last N months. However i've stuck with Claude because dealing with these "isms" and their little behavioral nuances is a chore in itself. I've found you have to learn the model just as much as anything, and so the idea of hopping these days when i'm just trying to get shit done is not likely.

These days for me personally, Claude has to give me a reason to switch rather than me investing even more money (i'm on the 20x plan) in other providers. I'm definitely not committed to Claude Code, but i am tired of the LLM churn, tooling churn, subscription churn, and the general fear of which providers we can trust.

edit: In short, it's the interactive UX just as much as it is the final output.

You're overestimating the extent to which individual developers have a choice here. My employer signed up for a Claude Code membership, I use Claude Code. I cannot use Codex.

Anecdotally I hear of folks with workplace Claude Code subscriptions all the time. I'm not sure I've ever heard someone talk about their workplace Codex subscription. Anthropic clearly did a far better job chasing corporate customers while OpenAI was busy chasing consumers with Sora etc.

Corporate accounts pay the full api price, so I don't know what is stopping them or you from also using codex on the same terms?

Intellectual property. My employer has an agreement that our code will never end up as part of Claude's training data. At this point there are also now custom Claude integrations etc.

I'm sure they could also negotiate a similar deal with OpenAI but in my outsider experience it seems that negotiations around these kind of corporate contracts takes forever and when the selling point is "they're broadly pretty similar" I suspect the motivation isn't there.

> My employer has an agreement that our code will never end up as part of Claude's training data.

“Our competitive advantage is that we believe them,” I’ve read—wonder if that’s still a [prevailing] sentiment.

(Edit - context was probably using SotA models instead of being limited to local open source only)

> My employer has an agreement that our code will never end up as part of Claude's training data.

Bit of a tangent but it is funny to me that so many companies talk about how some large percentage (sometimes 100%) of their new code is LLM-written and they still bother to worry about their code being used for training.

If an LLM is writing all your new code then your existing code certainly wasn't some unique special secret, and your new code came from the LLM to begin with.

I think the marketing campaign came first. Anthropic captured developer mindshare first, then they brought it to their companies.

Claude Code was a huge huge huge step up when it came out, absolutely massive.

It was barely marketed. I always turned copilot off, never found any benefit from Cursor. Claude Code was vastly different in conception, function, and capability, a product that defined an entirely new category of product.

Perhaps to others, that found copilot or cursor useful, it was merely marketing. But to me it was function and productivity, that I had never seen before.

People try to dismiss these things as LLM wrappers, but the LLM will be commoditized, and the wrapper will be where the real product design goes and where the real differentiation happens. Owning that unique process of communication between the dev and what the dev wants, figuring out the most stuff with the least complete spec, and maximizing every bit of the very tiny communication channel between the dev and the LLM and the code on disk, that's where 2026 and 2027 will be focused, until the next category defining product is created.

Sorry, no. Claude Code was the product that brought coding agents to the mainstream. Sonnet 3.5 was the model and harness that created vibe coding.

This was a push of the technical frontier, not a marketing achievement.

I started using Claude web with sonnet over chatgpt before any of the coding tools came out and noticed other founders were using it too and the reason was pretty simple - it was much less likely to hallucinate non existing APIs than ChatGPT

The OP seems unaware that Claude had a lead in this space and captured market share and attention for that reason alone.

The test they (supposedly) ran with their coworkers to look at PRs from both is such a bad way to compare LLMs that I don’t think they’re very experienced with using them.

> The OP seems unaware that Claude had a lead in this space

I remember using GitHub Copilot (OpenAI "Codex" mk1) in Aug 2021 (ChatGPT would launch a year later 2 weeks after Meta's botched Galactica release). Cursor & others took it and ran a mighty good race.

Okay so regardless of the model, platforms provide attestation from end to end that nothing is being logged from either input or output, including at the firmware and OS level to the extent that the customers have proof of the data never being saved. AFAIK, both GPUs and TPUs support this.

Claude had a lead for a few weeks. When Codex launched it was better and it has been (marginally, yes, but still) better since.

It's marketing

Having a lead in a new market even for a short time creates initial conditions that are in your favor. That’s why startups get all fired up over being first to market.

Sure, but OpenAI has caught up with such velocity (and frankly has the better models) such that it's kind of irrational to refuse it based on "vibes"

I know it is irrational vibes, but how the whole Department of Defense situation was handled will always make me partial to Anthropic. Unless there so huge shift, of course.

No, it wasn't and no one who used Codex and Claude since release could seriously make this claim. It was a magnitude behind Claude and only catched up recently

I have lots of choice (I own the company) but I'm still not going to switch from Claude until I see evidence that the alternative is meaningfully better. So far I don't see that evidence. In the past I've looked at using competitive products and it turned out to be a painful experience (Cursor didn't work at all on my computer, Google thing -- whatever it was at that time -- required dependencies I wasn't willing to install). I'm sure these issues have been resolved since but why would I spent time kicking the tires of another product just to have it work "as well"? Claude's cost to me is minimal so there's no cost savings to be made.

fwiw nobody "marketed to me". I picked Claude because friends were using it with great success and they helped me get started with suggestions on prompt style. Before that I'd played around with various LLMs for coding but not done any actual production work.

Maybe some of these companies will learn to stop appointing awful leadership then.

Having a sleazy CEO like Sam Altman or Elon Musk is a business risk. Many potential customers don’t like these people and they say abrasive and alienating things publicly.

Rolling over to the DoD’s desire for fully automated weaponry is more bad marketing. How many people switched from OpenAI to Anthropic over that? I sure did. Anthropic’s willingness to burn that bridge over an ethical stance said a lot about the company to me.

I’m not going to use OpenAI products for these reasons among others.

I’m also not going to use Cursor as xAI plans to acquire Cursor.

Maybe it’s foolish of me to avoid those companies for such petty reasons, but that’s not my problem. That’s their problem.

It takes years to build trust and hours to burn that trust to the ground. Customers can hold grudges for a lifetime.

This is especially true in a market with almost zero product differentiation.

I did a pair programming comparison over 3 month on Codex 5.2 and Claude Sonnet and my subjective experience was that based on cost and rollbacks to a previous commit Claude is significantly better. Especially in VS Code Copilot. I wrote a long Substack post about it. I would share its but its in the paywalled archive by now.

IME Claude has been a bit inferior. But, yeah, the marketing is just great.

Who doesn't think they are susceptible to marketing?

That seems like a strawman.

> Who doesn't think they are susceptible to marketing?

Lots of people. Yes, even on HN. Here’s just a couple of examples from a haphazard keyword search:

https://news.ycombinator.com/item?id=44787106

> Am I the only one immune to marketing then?

https://news.ycombinator.com/item?id=41186672

> I am immune to marketing

> We used the claude code and codex harness and I implemented some prs they needed with gpt5.5 and opus4.7 and asked them to identify which came from which only from the code.

> Couldn’t tell.

Why would you expect them to be able to recognize the signature of a model from a pair of PRs? I don’t understand why you think this is a useful test for anything when we have numerous benchmarks that run 100s of tests on models and both GPT-5.5 and Opus-4.8 perform similarly.

I have subscriptions to both. I run both on max reasoning. It is interesting to see the relative strengths and weaknesses of each model. You won’t always see it if you’re just scanning code. Some times one will spin for a long time on certain problems where the other has no problem finding the appropriate parts of the codebase and getting an efficient solution.

antirez made a comment that he and others found GPT-5.5 to be better at the optimization tasks he was working on than Opus. There are other classes of tasks where GPT-5.5 consistently stumbles where Opus will get a solution quicker. Lately I’ve been working on some code where neither model comes up with a good solution. That’s just how LLMs go.

The only reason you have seen more activity about Claude is that they got there first. Codex has been a step behind and GPT couldn’t match Opus at first. You’re testing them after they’ve closed the gap.

I am not sure why the past matters here. I am talking about now, it is a fast moving space.

As for the test, of course the output matters. Take image models for example. Differences are clear as day.

Should the fact that OpenAI existed before Anthropic did at all matter? No, imo. I would have used opus 4.8, but it only just came out- fast moving space

Correct output is table stakes. Your test only shows that the products work as advertised, it doesn’t reveal reasons why people prefer one to the other.

You’re guessing that it’s a result of advertising, and I agree that that’s probably a component, but it’s a mistake to assume that they are interchangeable when you have people saying to you directly “I use both and they’re not.”

If I were to give one carpenter a set of fine hand tools, another a full workshop with power tools, and they both made a picnic table to the same spec, and at the end I wasn't able to tell which came from which, would you say I have come to a fair metric for which type of tooling to use for wood working?

If the effort was the same as was in my test, yes.

Wouldn't the question be if they could tell the tables apart by quality (after insisting one of the two parties made things of superior quality)?

Yes

The amount of effort is an absolutely critical comparison, right? That's been left out, yet you keep on harping about how the outputs are the same, ignoring all the many many many comments that are talking about the amount and type of effort.

In fact, after seeing all these comments about the amount of effort, you redirected at calling that mere "vibes:

> Edit: i bet 99% of people here, if presented with a test where i gave 5 models but all of the results came from one, would not be able to discern this. Just vibes all the way down

Which, again, is a highly emotional way to view people trying to say that the process matters too. Calling people "vibes based" or "highly susceptible to marketting" and saying they take part in "tupperware parties" rather than evaluating their experience with tools is quite a thing to see, a complete dismissal of professionals' core experience as "vibes" rather than something intrinsic to how they perform labor.

This is an incredibly silly comparison. It amounts to claiming that a Ford Pinto is just as good of a car as a Rolls-Royce by simply observing that both cars got a person from point A to point B. After all, once someone reaches their destination you can hardly tell what vehicle they actually used to get there, but that doesn't mean there's no difference between vehicles.

What matters most in state of the art models isn't simply the final destination, it's the process of how one arrives to that destination.

In my test the prompt was the same and all suggestions were auto accepted so indeed there was no difference other than model and harness. The amount of characters typed and interaction with the harnesses were exactly the same.

To keep with the analogy, isn't that sort of like testing two cars by having them both drive the same few hundred foot stretch of new road at the posted speed limit of 35 MPH? You will test some things doing that, but not particularly well, and hardly all the things people find interesting and useful for comparing the performance of cars.

To bring ng this back to the discussion at hand (and to be redundant, as it's been mentioned here already), there are many aspects of using an LLM that are not purely about the output from a single or few well formed prompts. Additionally, if the end results are very similar, these othrr aspects will have an outsized influence on people's perspective of the tools, as they're the only differences worth choosing one model over another.

I think your analogy makes the opposite case better. A Rolls-Royce and a Pinto have the same real commute time because horsepower isn't the bottleneck, and they both get passengers from point to point. Sure the Pinto explodes a bit but much like the actuaries at Ford, you might well judge the cost of an occasional explosion to be a trade-off you can easily compensate for.

I would argue the process these days has more to do with the harness than the model, at least when we're talking about the SOTA options. Claude Code's biggest advantage isn't Opus, rather it's the shared knowledge the community has been building and sharing around using it effectively. Almost all of the out-of-the-box tutorials and skills and frameworks are build for Claude first, then Codex maybe.

I'd go further and say that CC and Codex are not even the best harnesses available, they just offer the most subsidized rate plans.

> Claude Code's biggest advantage isn't Opus, rather it's the shared knowledge the community has been building and sharing around using it effectively.

This. Never underestimate the ability of a large number of power users to substantially improve the actual utility of a complex software product.

They always have more time (and sometimes more skill) than a product's developers.

Sometimes the quantity of monkeys matters more than the quality of the typewriters.

To add on context, the experiment you're giving is called a *blind judging test*. Remove the branding and labels, and let judges sample the results and see if they can tell which is ranked correctly.

Some examples are blind wine tasting tests. There are instances whereby some journalists invited renowned/established wine tasters and subjected them to blind wine tasting tests. Turns out the judges couldn't tell which was which. Pretty embarrassing.

It speaks volumes as to how people can accurately judge the value of things. There is research by some network scientist that says you can't generally can't tell the 1% from the top, though you can tell the really bad from the generally good. What OP's experiment might tell us is that the LLM competitive advantage is so small no one can tell which is objectively better.

Exactly. Popular opinion is behind reality by several months. Claude used to be significantly better, now it is basically the same.

Claude has been behind since GPT-5. Claude Code just looked cool and had better marketing

Claude is more reliable in production, less errors and better understand instructions. That's why the valuation shifted, technical people are choosing Claude for actual shipped products.

This isn’t the case at all, the most technical and best engineers are all using Codex now and have been for roughly six months.

It’s a known “secret” for a while now how much better Codex is than Claude. I’ve used both since they were released and I often implement in both to compare and 95% of the time Codex writes better code and also less code!

Claude is only really better at front end design.

You're being silly. The actual technical people are using Claude for implementation and relying on MCP servers to use Codex 5.5 and Gemini 3.1 pro to build teams, councils, and long running senior engineer conversations within Claude to handle the technical bits that're too complicated for Claude.

How could you possibly know all the most "technical and best engineers?". Wait.. are you a codex instance?

Just follow respected folks on X who all stated they use Codex now over Claude.

You must not be talking about the Anthropic endpoint…

Kind of orthogonal to the discussion, but could you broadly describe the code you're working on that both models are bad at? One thing I'm still struggling with is figuring out what types of code LLMs can vs cannot write.

C code formally proven correct with Frama-C WP has been... marginal. The models do better than I expected at the proof portion (with ChatGPT 5.5 seeming to have a meaningful lead), but they all have a hard time (a) writing really good C code to begin with and (b) with compliance around not modifying C code semantics or performance as a cheat to simplify proof obligations. They also tend to be insanely and consistently verbose on the first proof pass... e.g. 8 lines of C code might end up at 200+ lines annotated and proven, but after simplification passes end up at 40 lines. I find I spend 90%+ of tokens on those simplification passes, and haven't really found a way to avoid the over-annotate-and-then-optimize tides by being a bit more sane the first time around.

> Kind of orthogonal to the discussion, but could you broadly describe the code you're working on that both models are bad at?

Commonly, anything that hasn't already been done across 100 different projects on GitHub.

Making a React app with a CRUD backend: LLMs are great. They've been trained on this.

Doing new work on complex non-public codebases or in niche problems that aren't commonly solved: Completely different story. Some times they'll find enough information to piece together a path toward a solution, but that doesn't mean it's a good solution. I also have to feed in a lot more context and even stop them when they go down bad paths frequently.

For the complex work I don't have the LLMs write code, but I may have them do a proof of concept. I have to write and understand everything myself. There are times when I'll think the LLM output looks good until I go through it line by line and realize it's done something completely unnecessary, or happened to get the right result for the wrong reasons. For unknown problems they're good at getting something to work through brute force if you let them consume enough tokens, but it may rely on safety fallbacks from the OS or fallbacks instead of being a proper solution. I always chuckle when they encounter intermittent errors and the first idea is to add a retry mechanism so the error is ignored.

Rules of thumb:

The more your toolchain (compilers, linters, etc) can statically verify, the better agents will do.

The terser the code, the better agents will do.

The more often similar problems have been solved in open source, the better agents will do. Agents seem particularly good at plumbing together different pieces of software.

Anything that requires a judgement call, as opposed to having one obvious way to do it, will get worse results from an agent.

As the scope of the request grows, agents get worse at it. This can be mitigated somewhat using various techniques ("write a plan", "do step 1 of the plan", etc) but never fully resolved. At some point the task is so big that it's necessary to do large parts by hand.

Yup, OP is conflating so many things that the comparison has all the scientific rigor of the Pepsi Challenge.

For a developer using an LLM on a daily basis, the experience is about much more than just the resultant code.

There’s everything from:

- how often you had to manually steer the model

- how frequently you needed to course-correct

- how much detail you had to provide up front

- how was the interaction process (sycophantic, etc)

- how well did it handle MCP and external tooling?

- how effectively could it pull in additional information from external sources such as the web?

- how fast did it produce code?

- how much did it cost?

Many of my friends who are devs use things like OpenCode CLI with Openrouter because they switch between the various SOTA models so often. Just because you saw a Claude "meetup" doesn't prove anything other than somebody chose the name because it resonated more than "Generic LLM Meetup".

the answer is : I usually let it do its thing with bypass permission and I run the max plan so nothing really matters except the "result". I think Claude is faster and has better UX integration with vscode but I wouldn't use it without GPT 5.5 XHigh as reviewer.Claude is just sloppy. Eventually I think it will not matter much in 1 -2 years. Most AI models will be good enough for most tasks so you may need the best of the best only if you do very complex stuff (i.e. optimizations etc)

I've actually settled on a very similar workflow - I mostly use Claude 4.6[1M] with adaptive reasoning disabled on High/Max for implementation, and then I'll do some combination of manual review in conjunction with GPT 5.5 xhigh.

I think the subscription pricing model kind of incentivizes developers (at least hobby developers) to pick one and go all in on it. For someone who has probably never paid $20/mo for a piece of software in their life, $20/mo is kind of a big commitment, and the pay-per-token schemes are reportedly much more expensive for the equivalent blob of coding they enable. So you "pick one," plonk down the $20, and use it as much as you can in the month so it's worth it. If you want to try the other one, you don't renew next month, and plonk down another $20 for the other one.

You can go back and forth and compare since you pay for both subscriptions, but is that a usual case? I'd guess most developers picked one in 2025 and haven't gone back. Just like most people just pick a bank for their checking account and never change it.

> For someone who has probably never paid $20/mo for a piece of software in their life, $20/mo is kind of a big commitment

I could see this being true for a high school student or college freshman eating rice and beans in their dorm room. Many of us have been there.

For someone in a software development career, $20/month for tools is a trivial expense.

I think some people have a strong aversion to paying for any tooling, but I don't think the people carrying around their $3000 MacBook Pros are going to avoid paying $20 for a month to try something new if they're using it daily.

Actually it would be fun to try to test the developer personality of the models.

Actually there is a nice body of work by Steven Clarke on cognitive dimensions of notations/APIs and the interaction with developer personalities.

I wonder if the same holds for AI models and harnesses.

> Some times one will spin for a long time on certain problems where the other has no problem finding the appropriate parts of the codebase and getting an efficient solution.

Surely this is just to the random nature of these stochastic parrots?

Do you mean you have identified a class of problems Claude always stalls on and another class of problems Codex always stalls on? What identifies these different classes of problems you see? How would you say Claude is stronger than Codex and vice versa? Why?

I just don't believe non-deterministic tools can actually be benchmarked. It's all hoopla to me.

I flip between models all the time. Makes little difference. Sometimes one model is faster or better than another but there's no rhyme or reason why.

All tools are non-deterministic on some reasonably specified input set.

> I just don't believe non-deterministic tools can actually be benchmarked. It's all hoopla to me.

We benchmark non-deterministic things all the time and it's frankly not even that unusual or hard. You yourself indicate that one model outperforms another one in your experience on various facets, and that is itself a benchmark.

The more relevant question is probably how well does a given benchmark translate to improvement on a specific desired outcome or task. The military uses the ASVAB testing battery to benchmark potential new recruits for suitability in various career specialties, but the actual outcome the benchmark is meant to correlate with is later success in the training pipeline.

So every so often the various military branches have to do and compare ASVAB results against training results and make sure that they still have a predictive relationship.

And this is benchmarking real flesh-and-blood human beings where you get on the order of magnitude of a million data points or so per year. You can benchmark AIs much more efficiently than that, as non-deterministic as they are, and as long as the benchmark itself is reasonably predictive of outcome it's going to be useful information.

My anecdotal experience isn't a benchmark. Just because I feel like something is better or different doesn't mean it actually is.

> Just because I feel like something is better or different doesn't mean it actually is.

Of course, but it is a data point, and multiple such data points can be aggregated. This is true even if all you can do is compare two things.

The shape of that data will reveal something more about the thing you want to measure than the null hypothesis you'd otherwise have.

The results are the same but I’ve found the process to get to the results are just more pleasant with Claude. I can’t put my finger on it. Overall most these models at the highest level are about the same in many respects but the UI/UX for some are just more enjoyable, for lack of a better term.

Codex I feel the need to be very specific and precise with. Claude… I feel like I can be lazy, which I enjoy.

Both still need to be reviewed stringently but I feel I can be more ambiguous with Claude and get better results than when Codex.

You confuse ease of using a tool with quality of output. A skilled carpenter can work both with high and with medium quality tools and prefer one over the other with no difference visible in the craft they produce.

Instead of only hanging them evaluate the final output, you ought to also have a way to have them evaluate the process and agentic aspects in getting to said output. Claude Code outshines when you look at it end-to-end, in my experience.

Benchmarking 1 or a few samples isn't ever going to yield anything but noise. The actual benchmarks use thousands of tasks.

GPT 5.5 genuinely was back on top for a while there, but if you look at the past 2 years, being on Claude was better than being on OpenAI most of the time. If you're going to pick a tool and not switch constantly it was the right choice. Not to mention their tooling has always been ahead, and that gets ecosystem benefits.

Are they close and interchangeable today? Sure. But Sonnet was genuinely way better than anything OpenAI offered for a long time -- the valuation reflects that, not any given moment in time.

okay what's a point in time where Claude was better? just give me a date

until GPT 5.4, Claude always had a decent edge in benchmarks (any date before then). The gap was huge during the Sonnet 3.x vs GPT 4.x eta.

100%

The belief structures here are really interesting. Blind tests would likely illuminate a lot of why people think that

Nothing new. I used to work with .NET and went to some meetups and conferences. There are some hardcore Microsoft fanboys out there. Didn’t even mix the kool-aid, ate it right from the packet. They only know MS products and seem scared of anything else.

Maybe not your typical HN crowd but marketing absolutely works on developers.

> Couldn’t tell.

add deepseek v4 to it, and it will be close at 1/10 th the price. I use all three codex, claude, and deepseek, and they are close.

If Claude meetups are the new Tupperware parties, OpenClaw meetups are something darker still. The tech is indeed worth discussion and celebration, but the Brands are clearly taking too much.

This is a really important insight. Great comment.

Agree wholeheartedly. I think that Anthropic has just invested more effort in creating a better DevEx than OpenAI, and so people just "feel" that claude code is better but they're about the same really, claude code might be 5% better at best.

A lot is changing. Like 9months ago, I was convinced Claude was best. I'm not so sure anymore :^)

Honestly I have no idea how you couldn’t tell. Reading a PR I can see the difference without even reading the words. (I doubt I could spot the difference just looking at the code diffs though.)

Claude commit messages - well structured test plan, readable.

Codex commit messages - wall of text, no structure.

The big difference though is sitting with the tools and using them for work. These are for sure vibes, but I’m sure you could pull out metrics for # steering re-prompts for example.

Codex just goes off and solves the problem, usually comes back with a solve; Claude more often gives up or needs input. Opus gives a broader design discussion, better at conversation. Codex finds deeper/better edge cases.

I think it’s like EMacs vs Vim - you can get your work done with both. There may be some tasks where one is way stronger. A strict “Better” is quite hard to justify.

Ultimately tool choice is a mix of science and art/taste; I want to feel joy using my tools, and fun little pixel explosions make me happy. If a different tool makes you happy, that is also fine.

I think frontier models are bait now. I prefer a fast, less thoughtful model that adhere's to instructions. The code these latest models produce is often hot garbage and you still need to micro-manage. Fast with small chunks is better.

I’m curious what models you prefer given this criteria.

I've been messing with DS4 flash. It's very fast and has 1m context. Downside is it's probably not great for private codebases (at least not when hosted by DeepSeek) and it's also not great at sticking to instructions if the task is too large / long though. I expect it will improve over time with better tooling and cache usage.

I basically have it load up a bunch of relevant context and give it small chunks of work in the same session over time (not like a fire and forget subagent). It's working fairly well. Bonus is I still feel like I'm part of the process instead of watching youtube videos while Opus / GPT vibe code a bunch of slop.

I mean, yes? Anthropic’s investors are seeing more upside now and valuing the company higher. Your thesis is that this additional value is driven by better marketing rather than a superior model. Could be! The truth is we’ll never really know with certainty what factors are doing the heavy lifting here, we can only guess and argue over who’s a better guesser.

You're delusional. Anthropic's success is not only the model - it's the harness. I'd absolutely be able to tell the difference between Claude Code and, say, Gemini CLI.

> I never want to hear from developers again that they are not susceptible to marketing.

Did you need to come to that conclusion?

Marketing has always been a significant part of new technology adoption. Whether it's for cloud adoption, for new programming languages, for new software development techniques, etc...

It seems we're moving past the point where it's all about model capability. opus4.7 behaves better for me than gpt5.5 because I'm familiar with its idiosyncrasies. Sounds like you've got a good balance between them.

At the end of the day what matters is which team is better, not which model. If Anthropic continues to feel like the good guy, relatively speaking, then people are gonna chose to spend more time getting to know its products and less time with OpenAPI's and on average Anthropic's will be the more capable teams.

I think vibes are gonna matter more and more going forward. The potential for bad behavior on the part of an AI company is severe. We're gonna have to tolerate whoever we enable in this space, so I propose that we make their marketing teams work as hard as possible to show us which will supply better vibes.

So you both used Anthropic models (Opus 4.7 being from Anthropic)? I'm struggling to understand what your comparison really was here.

I found that the newest opus and 5.5 are definitely close enough where most of the work I do could be done with either. I've seen small differences in planning which I feel like Claude does do better, but I think both products are close enough where I wouldn't be upset if one disappeared.

I find this pattern annoying and also commonplace: MY taste is correct. I AM right. The emergent properties of free market resulting from revealed preferences of free willed agents is WRONG.

Any name suitable to name this phenomenon?

The harness/UI that claude code brought was the thing that stole developer mindshare. Thats when people stopped coding in IDEs. Nothing to do with the underlying model.

It's a matter of what context is available to me at this time. I like LLMS. They improve my workflow to an insane degree. I think Sam Altman kind of sucks. I don't trust OpenAI. If they were the only kid on the block, I'd use Codex. It's entirely possible Anthropic sucks in the exact ways that OpenAI sucks but has better PR. I don't have time to deep dive to find out. I still like using LLMs. I started using Claude because Cursor, as a company, did something that I can't recall but gave me the ick. So I switched to Claude Code.

I still use Claude Code because I have the most experience with it now, and it's the harness that I understand on a granular level. If something comes along that is clearly better, or if it becomes clear the Codex is miles ahead, I'll try it and evaluate it. To your point, there doesn't seem to be much of a difference.

Arguing over this stuff feels kind of silly, like back in the day when my friends would give me shit for using mIRC instead of ircii or BitchX. I liked the GUI then because I did. I like Claude Code now because I do.

That sounds like someone desperately trying ton convince people Pesi is better than Coke

So you black boxed a few 'success' test, while the main diference between the two is the way they get to the result?

I've always interchangeably used the models.

I don't look at benchmarks.

It's a non-deterministic tool. A lot of the shit going on with LLMs just doesn't make sense to me. All the tooling around like MCPs, they're all just putting stuff into context. So to me the tools aren't really robust and they make little difference.

Lots of AI psychosis going on these days. And I say that as somebody that hasn't written a line of code since Sept 2025

Steam and other game stores are pretty much the same but Steam is more popular because every one of their competitors has decided to continually shoot themselves in the foot over and over.

Even if Claude and ChatGPT were exactly the same, Claude would be more popular because OpenAI has decided to make some very unpopular moves and try to make money where popularity isn't required. At the moment that popularity still seems to matter.

The funny thing about Tupperware is that some of us have their products from many many years ago and they still work great.

I think we've had the same iced tea pitcher since I was 5 years old, for example. Solid.

Will we be able to say the same thing about Claude?

> I never want to hear from developers again that they are not susceptible to marketing.

It’s a really good signal of self-awareness/arrogance

Newer GPT (5+) models seem to forget imports more often than Claude and use all lowercase comments more (possibly as part of OpenAI's effort to make it more concise).

It also seems to use modern Java features like var and records more.

I certainly can’t tell.

I honestly think I’d need weeks of all workday testing to even form an opinion… and some in depth training before that to use each given tool right…

And then … I might decide I can’t tell the difference.

As it is I use Claude and I don’t have the time to properly compare.

That comparison doesnt make much sense. The quality of a model isnt really its final output but the experience of working through a problem with it.

You sound upset at anthropic becoming bigger? Is this some kind of playstation/xbox fanboyism where people root for one company or another?

I like claude because it performs better in my experience, maybe ill check out codex to see if its good too, but i trust anthropic to have the sauce when it comes to coding.

The thing about developers claiming not to be susceptable to marketing is wierd. Im sure developers are susceptable to marketing and maybe there are some that claim to not be, but you seem to be overblowing that quite a bit, like anthropic ran a huge propaganda campaign to get claude out ahead. I think its not really conspiracy theory. Claude was just better at coding and the word spread among developers and people experience confirmed that. Maybe codex has caught up but people trust claude since its more established. Theres really no need to be mad about it unless youre invested in openai, financially or emotionally.

You attempted to create a deterministic test for an N-dimensional non-deterministic output

I don't think the success is due to marketing. They've been top of the LLM Arena leaderboard for most of the year which I think is blind AB testing. Most people on HN say they are best for code. I've never seen their marketing. Your post was the first time I even realised it exists.

I think Sam Altman is an asshole and I prefer to spend my money elsewhere.

Frontier models being commoditize is inevitable. OpenAI thinks they're still competing on technology, and not user experience and market reputation otherwise they'd understand the continuous negative PR generated by Altman's chaos is going to cost them everything.

He must have done something personally to you.

That's... that's not how social perception works at all.

Dario is genuinely as bad as Sam.

I’ve heard this said, but why?

he pushes mysticism of the models

he's starkly anti-China with a warlike posture that I find dangerous and unappealing

Anthropic has a much more confused mission statement than OpenAI

in interviews, Dario appears to care little for the well-being of common folk, while Sam at least pretends

I’d take honesty over deceitfulness any day of the week.

I guess I can see why you’d want to bury your head in the sand and pretend powerful people aren’t all that bad.

> in interviews, Dario appears to care little for the well-being of common folk, while Sam at least pretends

At least his ghostwriters do.

https://darioamodei.com/essay/machines-of-loving-grace https://www.darioamodei.com/essay/the-adolescence-of-technol...

I have yet to see anything out of Altman that indicates he's thinking about anything other than money or corporate PR.

Thanks for elaborating. Can you say more about mysticism?

I've come to dislike most tech CEOs at this point, and the current AI batch isn't making it any better. They rarely hold consistent beliefs, but nowadays are positioned as thought leaders.

I wish there was some type of system in-place to hold people to their word, but I can't imagine how it would work.

Altman does appear to be an asshole, but I have bad news for you if you think Anthropic are the good guys. If anything, they might be worse than OpenAI.

What makes you say that

Can you elaborate or give some examples as to why? I dont know much about this subject, last i heard, Anthropic declined deals with Military and government agencies - while OpenAI opened their arms. But i am not

I wouldn't say Anthropic is worse than OpenAI, but there's a lot wrong with them. https://anthropic.ml/ has a collection of incidents and relevant evidence.

Isn’t that site just loudly proclaiming that “pushing the frontier” of AI is inherently anti-social and that employees should be asking the public to “shut us all down” ?

That’s their core argument against Anthropic, that they are making progress at improving their models ?

Anthropic believes they are the only people who should control AI.

That alone is terrible. Also OpenAI has done actual work to try and mitigate the downstream negatives effects with UBI experiments.

Sam isn’t perfect but it’s deeply unclear that he’s worse

> Anthropic believes they are the only people who should control AI.

I’ve seen this a few times in the thread. Can you or anyone provide a link that supports this claim?

This seems to be a rumor being coordinated by OpenAI.

There's OpenAI employees spreading this rumor on Twitter with 0 evidence. Their entire evidence is "I keep hearing Anthropic wants to control AI". Their evidence is literal rumors.

https://x.com/_aidan_clark_/status/2052089187659346047

Yeah. And this thread seems so astroturfed. I get the feeling openAI is trying this sort of propaganda campaign out now that theyre desperate

What makes you think Anthropic might be worse than OpenAI? Anything specific, or just vibes?

Is there any argument to back that up or are you just hoping to self-actualize this belief using vibes alone?

How can you say this as if supporting Dario is any better.

At the top level of anything there is almost no such thing as a non-asshole.

None of them care genuinely about you they just want your money.

The world saw Anthropic take a possibly company-killing risk wrt weaponizing their AI, and are rewarding them for holding to their values, for now at least.

It’s not like anyone owes Sam Altman their business just bc their product has become slightly, perhaps temporarily, better

Both companies don't mind weaponization. The difference is whether it should only be domestic. Both are ok with foreigners being bombed.

That aligns with average human values

Do you have a citation for that?

Because the Anthropic's presentation of their position doesn't have a domestic or foreign caveat to autonomous weapons. It's a categorical no.

https://www.anthropic.com/news/statement-department-of-war

> We support the use of AI for lawful foreign intelligence and counterintelligence missions. But using these systems for mass domestic surveillance is incompatible with democratic values.

Your link

Yes. They draw that distinction on intelligence.

I was asking about autonomous weapons.

>> Both companies don't mind weaponization. The difference is whether it should only be domestic. Both are ok with foreigners being bombed.

Anthropic literally already works with companies like Palantir and others weaponizing their AI. Those just aren't quite as well known by the general public.

They have no values that align with humans prospering.

Source on this? Does anthropic work with them or do they simply use the normal product offering anthropic provides? Because the thing with the DoD was specific RnD for DoD purposes.

If someone just uses normal claude to make killer robots, i dont think thats on anthropic, that on the legal and regulatory system to step in and stop that

It literally takes a 2 second Google search of just the two terms Palantir and anthropic to find sources for this and is fairly well known.

https://investors.palantir.com/news-details/2024/Anthropic-a...

https://www.cnbc.com/2026/03/12/karp-palantir-anthropic-clau...

https://www.reuters.com/technology/palantir-faces-challenge-...

Any company of Palantir size is not just gonna have people using personal subscriptions to use Claude on mission critical software or it's development.

I hate that this discussion is about OpenAI vs. Anthropic and not OpenAI+Anthropic vs. Google.

Google put up so little of a fight against the DoW for their use of Gemini that we didn't even hear about it. They are clearly the worst of the evils here, but OpenAI is the one getting all of the negative press.

if dario is just as much of an asshole, he's at least a quieter asshole. and to me, that's better.

"Silent but deadly"

Whereas sam altman is "All fart, no poop"

> At the top level of anything there is almost no such thing as a non-asshole.

There's only one Gabe.

> None of them care genuinely about you they just want your money.

It's worse than this. Billionaire entrepreneurs aren't funds manager, they don't just want money, they have a twisted sense of “being the good guy” driving humanity forward against its will.

> driving humanity forward against its will.

I want this technology! You don't speak for all of us.

I'm sick of techo-Luddism. I'm sick of complaints about water use in a world that has avocados, beef, and fabric dyes. I'm sick of complaints about power use when you have your air conditioners, winter heat, air travel, and gaming PCs.

I'm sick of artists saying AI image and video sucks. I'm sick of pretend artists, armchair warriors, obsessed fans, and pickmes towing the same line.

I'm sick of engineers saying these models aren't a huge performance gain.

You haters and skeptics out there can keep doing you, but I'm going to keep using the technology. We'll see where the chips fall.

I was raised on optimism and dreams of the future. I want that. I don't want to die with the same incrementalism we've always had. I want orders of magnitude more.

This is our one moment of awareness in a cosmically infinite void. I want spectacular. I'm tired of the chicken attitudes when people should aspire to be eagles.

What we have is so boring. There's so much more if we reach for it. Holodecks, models that cure every molecular cause of cancer, doubled and tripled health spans, instant ability to understand every language, fast and cheap travel autonomous p2p travel, everyone on earth lifted out of poverty, a Michelin chef in your kitchen, ...

None of these billionaires are going to lead you to the dreams of the future that you have.

And don't forget the negative effects on other people, otherwise that's just selfish.

It makes me sad to read someone rejecting humanity.

Painting something new with great potential as negative and "the enemy" - this is truly the essence of humanity. </snark-but-with-a-kernel-of-truth>

If the parent commenter enjoys working with LLMs, then just let them. This doesn't make them inhumane, nor does it make you less of a human.

Putting the AI in cocaine, eh?

That’s a take. Would be cool if you could be happy with what is, rather than what you wish could be, while still being optimistic about new tech. But even if you can’t, would be nice if you could have your optimism without letting other people’s differing opinions get under your skin.

I’d also suggest, if you care about things like curing “every molecular cause of cancer” to spend some time and energy working in that field to understand the real problems there and work towards real solutions (with models or whatever floats your boat), rather than hoping that some poorly defined techno-optimism hand-waving will just happen to result in the best of all possible worlds, with no downsides or alternative outcomes.

Also, crazy to say that the miracle of existence is boring because we don’t have the tech you imagine!! If that’s your take now, no new technology is going to fix it. You’ll just still be bored with the holodeck and your one precious life to live.

There’s a difference between being a hater and acknowledging the reality of the technology and those building it. I want all of those things too. However I do not understand why LLMs will get them for us. Instead I see a few really powerful people looking to get more powerful. I see a powerful tool being presented as god in a box when. I see the most resources ever spent on a singular thing being spent in a way that’ll _best case_ be mostly obsolete in a few years.

I want real AI. I want cures for cancer. I want too want to live in a post scarcity world. We had most of the technologies to do that before this. However the companies and investors involved in the AI build out chose to sit on massive reserves instead of trying to directly solve those problems. There exist proposals which solve hunger, the energy transition, etc and together they wouldn’t amount for even half of what’s been spent.

That tells me those involved want nothing other than money and power.

Reminds me of Mike Prince in Billions. God that guy was insufferable, just like the rest of today's broligarchy.

I thought that show went downhill once Bobby Axelrod left but they did a good job with making Mike Prince a huge ahole.

At the top level of anything there is almost no such thing as a non-asshole.

What a sad, bitter worldview. I hope you find some peace.

You have a point but I think you might underestimate how much it takes to be a snake like Sam Altman

Not that I have first-hand knowledge but if reports about him are only half true, most tech CEOs are already saints compared to him

Well anthropic refused to do shady shit with the DoD and openAI did. So theres that. Also altman wants a private global ID system that screams technofascism.

This comment just sounds like baseless both side-ism. Anthropic arent saints but i dont see why they have to be. They are business' and one is less trustworthy than the other. Simple as

Hassabis seems decent. People have strange attitudes to tech leaders like they are all superhuman or evil or all aresholes. They are just normal people who got the top job, some good some bad, some mediocre.

I think all US companies are turning extremely anti consumer hostile thanks to unchecked unregulated capitalist greed and I would prefer my money to the Chinese underdogs. Cannot speak highly enough of Deepseek.

The idea that intelligence will be commoditized is completely counterintuitive. It comes from the belief that it can’t exceed our own. This is almost certainly not true at the limit. There will likely by many super intelligences like we see life in the wild

No, it comes from the idea that the intelligence each company offers at any time will be undistinguishable. Sure some models will temporarily pull ahead, but others will quickly catch up and the intelligence difference won’t matter enough too convince anyone to switch on its own.

That is an incredibly limited an shortsighted view of AI

Maybe. Its been pretty true so far though. A singularity is possible of course, but it's also possible that intelligence explosions dont happen, progress is relatively linear, and no one is able to pull ahead long term. If the true bottleneck is hardware it's difficult for me to see how intelligence doesnt become a commodity.

OpenAI and Anthropic are the same shit. That people think that Anthropic is in any way morally superior is sort of laughable.

>I think Sam Altman is an asshole and I prefer to spend my money elsewhere.

That's why he chose the OpenAI logo

And the Anthropic logo is a subtle reminder to stuff that ejected out of OpenAI.

This is exactly it!

Sam Altman is the main perception problem for OpenAI. His background, history, trustworthiness, vibes/interviews etc are all negative PR when seen by the common man.

Dario is more knowledgeable, well informed, empathetic w.r.t. problems etc. In short, somebody who seems mature and trustworthy.

Sam and Dario are both assholes of the highest order. Sam just has more masks, and is better at wearing them.

Is he? I don’t remember anyone throwing a Molotov cocktail at Dario’s house

This is an absolute joke.

Anthropic capitalized upon a brief window of being more code-focused, which turned into enterprize contracts.

Then on renewal rug-pulled those same enterprises - going from your seat includes all the usage a user would reasonably need, to being you pay for the seat + all tokens at API pricing. (Which they raised by how many times in a year? I don't know the actual number.)

Revenue spikes like crazy through basically hostage taking made possible by Sonnet 3.5 era sentiment + enterprise purchasing lag.

Parlay the revenue spike into the valuation.

Crazy. Those same enterprises will get sticker shock and leave. Absurd short-term thinking.

OpenAI is the better company (transparency, open sourcing things, how they handle things in general e.g. OpenClaw, how they compete, etc.) and they have the vastly better brand, the better consumer presence, and (for me and many others) they have the better coding app + models.

Anthropic doing deeply customer hostile stuff - again and again - to produce a short term revenue spike does NOT make for a long-term sustainable business.

For such a young business to have such a long history of bait-and-switch is absolutely crazy. (Raising prices repeatedly, lowering rate-limits repeatedly, changing the terms, banning calls which contain "OpenClaw", turning on their IDE partners, turning on their enterprise partners.)

AFAICT anyone who's ever shown faith in Anthropic has been immediately exploited by them to some degree. They will quickly get the reputation of being "the Oracle of AI companies".

I wouldn't even value them at half of OpenAI.

I somewhat disagree, there was a major, major shift in developer sentiment towards agentic development starting with Opus 4.5 in October. Many teams started finding a lot more real value in Anthropic than they used to very recently. Things like OpenClaw are not a part of serious enterprises yet due to the security risk.

I’m not sure how anyone believes that per-seat pricing is halfway viable for AI, and I’m fairly sure the organizations I’m familiar with only REALLY started committing to spend after the shift to API pricing, due to the value they thought they were getting anyways.

Anthropic nuked a big chunk of that "developer sentiment" when they rug-pulled us with the rate limits and gaslit us with "it was just a bug, guys!".

OpenAI insider found!

Anthropic hasn't done anything customer hostile to me.

Notable and persistent and extraordinarily petty is their refusal to read AGENTS.md files, forcing the inclusion of their branding into the source of repos directly.

That is actually pretty wild to me. Do you have any sources regarding this?

I see my GPT5.x harness seek out and read CLAUDE.md files all the time.

Like how safari and chrome have mozilla in ther userstring. History does rhyme doesnt it

https://github.com/anthropics/claude-code/issues/6235

to be clear, I don't mean that Claude refuses to read AGENTS.md files, but that Anthropic refuses to bake AGENTS.md into the harness as a first-class feature.

Claude the model by now has learned what an AGENTS file is, and itself is not petty enough to ignore it for partisan purposes, but claude-code the harness doesn't natively support it.

> Crazy. Those same enterprises will get sticker shock and leave.

They are already. Both for sticker shock, and also because of developer sentiment beginning to shift towards Codex. …and then in a month or two the winds will shift again, I'm sure.

It's interesting to see how Claude Code got commoditized so quickly.

---

> and they have the vastly better brand

Strong disagree there. Anthropic has pretty successfully branded themselves as the more ethical & 'human' of the two companies. (whether that's the actual reality is irrelevant)

My dude, this level of vitriol and hyperbole feels excessive. It's okay if you don't like Anthropic and you disagree with their business practices, but this is over-the-top even for the internet.

Sam Altman appears to represent a significant liability for OpenAI’s success from this point forward. A big portion of the driver for Anthropic’s meteoric rise over the last six months appears to be folks recognizing “it’s that AI startup not run by Sam Altman.” Anthropic has amazing tech, but its biggest asset at the moment seems to be that “it’s not OpenAI.”

Not saying that’s right or wrong, but it’s clearly a factor holding OpenAI back at this point.

What's incredible is that Anthropic are clearly not saints but Altman makes them look like the good guys, reinforcing their marketing.

When Anthropic had the dispute with the Department of War over very meek conditions (a truly moral AI company would not be engaged in war crimes in the first place), it was a test for Altman, all he had to do was to take the same position. But because he's a psychopath he failed that very basic test.

I don’t know that most people care at all or even know about this. ChatGPT still far and away has the largest consumer market and brand recognition.

What Anthropic has done exceedingly well is work their way into corporations.

I have personally seen massive uptake over the last 6 months of regular people in corporations using Claude cowork. They are all genuinely amazed by what it can do for them.

OpenAI wants to be more of a Google. It’s increasingly seeming like consumer may not be as good of a play here

Even in the consumer space the cool kids stopped using ChatGPT this year. The swing in reputation and momentum these last few months has been nothing short of extraordinary. Like a political race, once a leader loses “the big mo” it’s incredibly hard to win it back. That’s the situation OpenAI faces now.

Allegedly OpenAI's contracting model is much more vicious than Anthropic's; at work (admittedly a little IP-protective) we have unlimited Claude, but no Codex subscription because OpenAI won't give us sufficient guarantees around data retention.

We are also concerned that it may not be possible to bind OpenAI using contract terms and/or the US legal system.

> OpenAI wants to be more of a Google. It’s increasingly seeming like consumer may not be as good of a play here

OpenAI has openings right now for "AI Deployment Engineer"-style positions, which is a role where they embed that employee in one or more customer's businesses. E.g.:

https://openai.com/careers/ai-deployment-engineer-startups-s...

I think this is the right way to go about it. Getting AI integrated well is more of a consulting package than it is a technology/code thing. Just handing a business a model+API will not result in high-quality or long-term relationships. This AI transformation is the most invasive possible thing I can imagine for a business. You really need a human on site to help the other humans across the treacherous organizational and psychological bridges.

My short experience with Claude and ChatGPT via web is that:

1. the way GPT writes is simply fundamentally annoying. I pretty much had to create a project with a file that said "do not use headings, lists or emojis" to make it bearable. It feels like, as a product, this sort of thing should be a general preference the user sets before they even start talking to a chatbot.

2. Claude just loves wasting tokens doing things nobody asked for. You ask "how do I calculate the distance between 2 points?" and it's probably going to compile some C code in the background with tests to make sure it works, then generate an interactive diagram on the fly to show how the math works, and then give you a downloadable file with the code. Like, dude, I just want some text. Why are you doing all of this?

Both of these problems come from the obvious lack of any UI controls in the software. there is no way for the user to know what sorts of things the software can do, because it's not exposed via UI as a checkbox like "generate interactive diagram" or "avoid using emojis." Discoverability is burning tokens to figure out what prompts work, or looking at example prompts the developer placed in the welcome screen.

I just feel it's completely ridiculous how LLM's are essentially the culmination of a trajectory of bad UI practices masquerading as "good UX" and now they're being implemented everywhere because people think it's good UX a blank textbox where you don't even know what you're supposed to type to do something.

I don't know, have you ever watched a video of Dario Amodei talking? I dislike them both immensely, but Dario both seems to take glee in scaring the shit out of people as a hobby with careless statements and his voice sounds like an impression of Elizabeth Holmes (the weird fake deep voice), and nothing he's said or done makes him seem more trustworthy.

I have never seen a video of Dario Asmodei. Which tells me he's less of a liability than Altman. For now.

Haven't seen much, but the things I've seen from Altman directly give me uncomfortable feelings, and the things I've seen of Amodei seem reasonable, even interesting at times.

OpenAI’s models could be materially better than Anthropic’s and I still wouldn’t use them because I don’t want to support Altman.

Do you think Amodei is different?

The choice is not binary. I use DeepSeek (paid) for coding, and Qwen (free) for casual stuff from the browser chat UI.

I mean paying for deepseek is sending money to China, and any company in China is pretty much an extension of the CCP (or can become one at the snap of their fingers). I don't think this is much better or worse than the American AI companies.

I mean, Deepseek is open model, it can be rehosted in the US by anyone, there’s no royalty to china

Deepseek proper is the only provider offering Deepseek V4 Pro at such an insanely low price. All the other providers are like 5x.

If someone says they're using Deepseek, it's probably a safe bet that they're sending money to China. People choose an inferior model (Deepseek) because it's cheaper. Not many people are willing to use Deepseek but also pay the 5x price to have an American company run it for them.

I run Deepseek at home...only people getting money is my electricity company

It's a testament to how much American companies have been fucking over their own reputations that people would view them as not "better or worse" than an extension of the CCP.

A decade ago, just that kind of comparison would've been pretty unthinkable.

Deepseek is much cheaper so that means you spend less on CCP than on Anthropic/OpenAI. Then there are third party providers as well.

>...sending money to China...

Have you seen the trade deficit lately?

If that's a concern, an easy alternative is using companies like novita or fireworks.ai to run open source chinese models. They are incredibly price competitive, have strict data retention policies, and in general are quite fast. Also, you don't get used to a model and suddenly get rug pulled when the provider decide to upgrade.

you can use deepseek without giving china a cent if you want.

Wait, people actually pay for DeepSeek?

The $20/month Claude Code plan has incredibly low rate limits. $20/month of DeepSeek usage buys you a lot. Even more so in hobby projects where usage is bursty, not nicely distributed over a work week

Hm, Altman vs China. Peculiar choice.

And honestly, DeepSeek is an excellent model.

I am having a better experience using DeepSeek than using Anthropic.

Not binary, but the parent was talking about OpenAI vs Anthropic.

No. He's worse. Much worse.

Why?

Why is Altman worse?

Dario is constantly fearmongering to generate press, gaslighting, and contradicting himself. Mythos is the most recent example of that. It was never too powerful to release, that was a lie to generate publicity and fear, and an excuse because they didn't have the compute to serve it. People were finding the same bugs and exploits using GPT5.4, GPT5.5, and lesser models. Now all of a sudden, they do have the compute, and now they're saying that Mythos is releasing in the coming weeks.

Anthropic is constantly caught up in ethical scandals too. They pump the web full of advertising bots. They steal peoples tokens, punish you for disabling telemetry, blacklist people they don't like. They had remote code execution vulns in their product for nearly a year and secretly buried that fact, no disclosures at all. Here are some of them https://clawd.rip

They're the least generous with open-source. The most closed off. The most likely to punish you for doing something they disagree with. Whenever OpenAI has issues they reset Codex's rate limits, they've done this every month that I can remember, and sometimes several weeks in a month. When's the last time Anthropic has done that for the many service issues they have had? Never. Not once.

Anthropic also never reply to peoples complaints or issues on GH issues, meanwhile the Codex team is very responsive and they actually care about customers user experience.

There's more, but you get the point. And yes, obviously not all of this is about Dario himself, but he drives the culture at the company.

>They're the least generous with open-source.

Yes I'm surprised most of the "anthropic stans" don't realize this.

It's not like OpenAI is a particular fan of open-source models either, but the way I see it to OpenAI it's a purely calculated pragmatic business decision. They won't release weights for any good models since it'd impact revenue, and they view other open models as competition. But otherwise they're not ideological about it. If there was some really good reason to, they would (and they have, gpt-oss probably as a fig leaf). Sure some of the characterizations of "sama" aren't particularly flattering, but at the same time I think the common thread that he views everything as "just ruthless business" is ultimately beneficial, since rational and pragmatic can be modeled and understood.

Anthropic's founding philosophy is built on the premise that only they should control AI because of its danger, only they can be the right stewards, and that open-source models are an existential risk. Oh and the fact that it's Chinese open-source models rather than western models that are "winning" must be keeping them up at night.

Compared to Scam Altman? Infinitely

Amodei is convinced he's abraham prefiguring AI's christ. Very different than Altman's cold power-seeking. You can always trust someone who's selfish, since they'll always do whatever's in their benefit at all times.

Is that better? I'd much rather have someone trying to build a good product to sell to willing buyers at a fair price than some zealot who thinks he's going to usher in the second coming and end the world or something.

Yes

Do you hold any amount of power in the world? A project that people care about, or a deliverable that someone depends on?

Just curious how you can afford to care about the guy 7 levels above the men that built and support the API that you buy.

Some people care about things beyond their own immediate self interest.

Some don't, and find it hard to believe others really do.

It’s also a weird argument. You can only spend your money once, and the affected employees also chose to work for a bell-end like Altman (or Zuck, or Musk)

I enough 'small' senior developers, project managers, product owners, internal IT people take a small stand against OpenAI products, that can still sum up to a notable impact

What is this Sam’s alt account?

People can spend money how they wish. SamA is a prick, so I don’t buy from his company. I don’t buy from Microsoft or Oracle either. Giving a company your money is explicitly supporting them and everything they do. Are you going to force me to buy products from people I don’t agree with?

Being a “prick” is completely inconsequential compared to the massive harm that anthropics views could create.

Such as them genuinely believing they are the only ones who should control AI. What could possibly go wrong?

I think it's about whether you trust someone or not. I don't really find Altman trustable. No one should have sole control over AI, but we need to have some trust in the people that are operating it.

why would you spend even a fraction of a second defending him

I heard my kids argue last night. My daddy is so tall. My daddy is bigger than the house. No my daddy is bigger than a roller coaster. Yeah? Well my daddy is 50km. You mean he is long? Yes, my daddy is longer than you.

I was cracking up. I'm 5'7 on a good day. I feel like that's how valuation works. We are propping up five foot tall giants.

Only on a good day?

Low gravity day

Unicorns, strapped with rockets, too busy looking at each other to realise the Earth is far gone.

They'll kill us all, or they'll kill each other. They sure as hell ain't making the world a better place, like they promised.

Dario really gives: "I'll make the world a better place after I burn it to the ground, I promise."

Claude has already been implicated in the triple tap strike that killed over 150 school girls.

They used Palantir's Maven to identify and prioritize targets and Maven integrates Claude into its decision making.

In any sane world where war crimes by the US were actually being taken seriously, both of these companies would be sanctioned.

Ah, it’s a good time to check in with gwern on our conversation about oAI vs Anthropic: https://news.ycombinator.com/item?id=40816755 and our predictions (ca two years ago).

Upshot - poetry expertise does not seem to be the primary focus these days, perhaps to the detriment of the entire world. We did move on from training scaling to “test time” scaling (which I hate as a name btw), Ilya does not seem to have been needed, (although I am really curious what he’s building).

My prediction that you want to be deeply embedded and really rich and part of global infrastructure feels good. My suggestion that oAI / MS would be able to use the lead in 2024 to extend was wrong.

Neither of us talked much about coding as a product that would drive value and behavior, which is super interesting to me, we were probably six months from seeing real competence of any sort there way back in June 2024.

We both seemed to think there would be a single breakout company, or could be one, (although I did suggest buying the basket), clearly not the case with GOOG oAI and Anthropic all posting serious revenues this last quarter / year.

One area of Anthropic that was nascent in 2024, but that I have come to think is super valuable is their mechinterp group. I still don’t see work done by other labs (at least published) to nearly the quality of Anthropic. And the group has clearly moved into a period of productivity; there’s a good chance in my mind it could provide a truly enduring strategic advantage as a tool to be used by the taste makers steering the ship. In 2024, interpretability seemed almost impossible to get a handle on — today, the sustained chipping away at the problem makes a lot more look possible.

Mechinterp in general is just completely undervalued right now (and agreed Anthropic's team is doing the most rigorous work, now accompanied by Goodfire). They're doing the closest work to neuroscience's in vivo 'thought-tracing', which is just the most wild science fiction sort of thing to be working on, and yet I feel the average person has no idea this sort of work is happening. When combined with the idea of the 'universal subspace hypothesis' (explored under the paper of the same name), you really start to bridge the gap from engineering to something more philosophical and spiritual. But I digress...

Haven't heard about the universal subspace hypothesis yet, so I appreciate the digression.

Ya, super interesting research area the authors explored of basically trying to answer the question: "Is there a canonical/intrinsic way that concepts/representations/information are 'stored' in the universe/reality?".

They tested that by performing "spectral analysis of over 1100 models - including 500 Mistral-7B LoRAs, 500 Vision Transformers, and 50 LLaMA-8B models ... by applying spectral decomposition techniques to the weight matrices of various architectures", and concluding that "deep neural networks trained across diverse tasks exhibit remarkably similar low-dimensional parametric subspaces", showing that "neural networks systematically converge to shared spectral subspaces regardless of initialization, task, or domain".

Not just philosophically interesting but also has practical implications for being smarter about how to reuse models, model merging, developing more sustainable training and inference algos, etc.

Paper source: https://arxiv.org/abs/2512.05117

Yeah that layer looping is super interesting. Also, it’s memory friendly with the right inference harness.

Very Chomsky friendly. Interesting paper, thank you!

it's not undervalues, many people are working on it following anthropic's lead. It just doesn't seem to be any useful, so it's even overvalued

Did you also talk about "head and shoulders" and "pennant" patterns in stock charts? Or where the "smart money" is at? I'd like to subscribe to your paid newsletter.

This is a super low quality comment. I try to put my current thoughts out here largely because high quality comments refine my thinking.

What about the original conversation seems correct / incorrect / interestingly correct or incorrect to you? What about my summary now seems correct/incorrect, etc?

Ultimately the big value is to me — I get to look back at some dialogue and then check where I had it right, wrong and interestingly wrong. My personal hypothesis is that doing this a lot compounds in a good way.

At this point I think it’s more important to have a solid workflow and understanding of how [insert your favorite model here] works and its capabilities, than chasing the next shinny release jumping back and forth between companies. I just finished my first large project with Codex and it is hard for me to believe Claude can be much better. It may be a bit better or worse, but again, they are all so good now that the user is the one driving the difference.

Yeah none of this is new if you've been working in this industry for a while

We've always been having these debates over whether my choice of tech is better than your choice of tech, same holy wars, different type of tech

The advice today is the same as it was 10/20/30+ years ago, pick what works for you and build something good with it

Nobody will actually care how you built it, regardless of whether its good or crap (although if it's crap you can blame your tools)

> jumping back and forth between companies

There is currently quite strong incentive to establish vendor lock-in for Anthropic and OpenAI. I see the ability to jump between companies quite important, especially for larger users. Right now it should not be hard but it can be much harder in the future.

codex gtp-5.5 is far superior to opus 4.7 working on large projects

Not everyone is a developer...

And 4.7 is so last week..

Soon none of us will be! right?

GPT-5.5 is the better programmer but Opus 4.8 remains the better system architect and product designer.

Codex is very "miss the forest for the trees", but is much better at successfully making large changes in large codebases. Claude Code makes more mistakes, but has more taste and a better grasp on idiomatic and elegant software development.

If you can afford to, I recommend juggling both.

I find arguing that a complex weighted graph has a taste is interesting.

This is not a jab, but a genuine curiosity of mine.

The taste that the complex weighted graph was trained on was better for one than the other I think is the long winded way to say it

The roulette pockets for the model are bigger for some outputs than others. Draw a big enough black box around it and a different one around humans and it's insistinguishable.

More interesting than arguing a jumble of electrochemical reactions have taste? That may seem more readily familiar but is no less strange if you prod at it. Nonetheless it’s difficult to argue either don’t produce output that has qualities of discernment (ie taste).

Isn't it just arguing that one complex weighted graph was tuned to output tokens that more align with what current day users would define as 'taste'?

I don't think it necessarily says anything about a model itself having 'taste' in some subjective way.

If the fashion changes would the model update with it without retraining? No. So the model doesn't have 'taste' in that sense. It has alignment to current human definitions of taste.

It is more capable of writing code I find tasteful and maintainable. It is debatable to what extent it itself has taste. Its outputs just suit my taste more than Codex's do, even though Codex introduces fewer bugs.

Great analysis and follows my experience as well. Codex is better when you know how you want the design and the architecture and you drive the agent a lot more aggressively. Claude Code feels like more autopilot so executives and users who didn’t code before AI like it a lot more.

But I feel like an expert who can drive GPT aggressively will out perform Opus. It’s why some smart people I know are opting for GPT and have fallen off on Opus. It’s like asking an F1 driver to sit in a taxi.

This is exactly right. Claude has baked in autonomy and preferences that let it handle underspecified prompts elegantly, which makes it seem smarter to people who like to prompt that way, but it also ignores instructions and fights you on things, which makes it a bad model for people who know what they want to do and specify it.

Opus 4.7 (haven't tried 4.8) just really struggles writing correct code for complicated (i.e. valuable) work. I can handle architecture, which takes <1% of my time anyway. But writing code that's wrong is a cardinal sin. I've had much more luck with GPT 5.5 so far.

In what ways? LM Arena has Opus 4.7 w/ 1567 -/+ 7 vs. 1505 -/+ 10 from GPT-5.5 Codex in code. I'm currently using both.

Admittedly my recent experience tilts Opus now 4.8, but you and others have my interest piqued re: GPT-5.5 Codex so I'm trying that more now.

arena is not a good benchmark, it is very susceptible to sycophancy

Opus 4.7 is not the current version of Opus.

GPT 5.5 still invents facts rather than looking them up, and manages to come across both as condescending and sycophantic. It feels like talking to a used car salesman.

Funny cause I'm quite literally having this exact issue with 4.8 as we speak. I've been going back and forth with Claude since yesterday afternoon on chopping up, stabilizing and facilitating recovery on a flaky mega-pipeline. Not 5 minutes ago, I had to remind it that two of the solutions it proposed were not possible because the target technology doesn't allow what it wanted to do, despite pointing it to the very docs that says it can't be done in the first place.

As far as its tone... Both feel like sycophantic as hell to me. To be honest, they just all feel so.

> GPT 5.5 still invents facts rather than looking them up

So does Claude, what’s your point?

I used it and ChatGPT this week in trying to assist troubleshooting a complex DB related issue and Claude had to apologise no less than three times in which it admitted to talking complete shit.

Just one example of the kind of shit it dribbled:

> I need to be upfront with you. I should not have claimed X as if I knew that for a fact. That was overreach on my part.

I've noticed that Claude has made less mistakes than in the past. I feel it checks it's own work more rigorously now, and understands it's own claims better and knows how to confirm them.

It rarely happens to me that Claude comes with clearly wrong modifications. Only with quite complicated problems with unclear variable names for example. But usually Claude asks me when something is unclear.

No experience with Codex though

You're using last week's model; Opus 4.7 is old news. Opus 6.9 is the new hotness; it is a better product manager than GPT, and has more X productivity. It replaced our junior dev team, and tells me my hair looks good.

Your research finding LLMs ineffective is invalid because you used 6.9. The current SOTA is 6.91 and it's leaps and bounds better that yesterday's 6.9

Fuck; you are right.

I'm experiencing the same. Codex gtp-5.5 has more brilliant intuitions, write less code, i.e. it identifies the exact point in which the modification shall be done. Nevertheless, huge improvements on personality from opus 4.7 (it was too accomodating) to opus 4.8

I strongly believe the reason gpt-5.x performs so well on large projects is because of the focused training they've done on their dedicated apply_patch primitive.

The official implementation of apply_patch is well thought out. It is a two-phase process that will not actually make any changes until all files in the change set are not ambiguous. The pre-commit error feedback usually fixes anchoring issues with one or two additional attempts. It generally goes something like:

  Reading file A L1:154
  Reading file B L1:123
  Attempting to apply patch... 
  [anchor errors for both A & B]
  Reading file A L43:67
  Reading file B L50:74
  Attempting to apply patch... 
  Patch succeeded! Running compilation & unit tests...

The anchor error feedback helps massively because in this implementation it also returns the current line numbers where the problem was found.

Techniques that replace the whole file or depend on find-replace are useful in more isolated contexts. However, when you need to refactor 20+ files, something like apply_patch is what you want. Anything that depends on specific line numbers for actual replacement targets is a total dead end for complex edit scenarios.

https://developers.openai.com/api/docs/guides/tools-apply-pa...

My problem with codex/gpt that is too verbose (mostly js and python): a lot of helper functions, a lot of 1 or 2 line functions used in 1 place only, a lot of types or proxy like objects.

I have specific skills for trying to avoid this, but nevertheless I spent half of the time fighting with its verbosity.

Currently, I'm trying to scaffold the functions/classes I know I need with NotImpelmented and ask it to implement only inside those specific places. It's a little bit better, but I still have to fight with function in functions definitions ...

My experience as well. Although this week I've moved to Cursor and Composer 2.5. It's so fast that any faults can be iterated on super quickly. The model is just insanely good with code things.

source?

In this game, who wins - in the long term - is who has the best model: so far OpenAI is ahead, so in the long term this is what matters. However, for the same reason, if in the future open weight models will be very near the quality of frontier labs, Anthropic and OpenAI will be out of business very soon. The game they play only make sense if their SOTA models do things that other models can't do at a comparable level.

IMO bad take.

You can theoretically do most things AWS does most of the time, yet people pay premium for it and keep paying for it, even though alternatives are cheaper, simpler and more performant.

I'd bet you that after 20 years OpenAI and Anthropic would still be around and kicking.

You might have a subpar product (for the price) but the reputation and history is what makes people open their wallets.

> I'd bet you that after 20 years OpenAI and Anthropic would still be around and kicking.

Depends. The bigger the bubble, the bigger the pop.

Only a few unicorns from the dot-com bust came out the other side (Amazon, Google, ... anyone else?), and that was a piddling affair compared to this one.

Yahoo is still around and kicking. Even Lycos' corpse is still warm.

> You can theoretically do most things AWS does most of the time, yet people pay premium for it and keep paying for it, even though alternatives are cheaper, simpler and more performant

It's going to be debated forever whether wiring your own open source tech has a lower development cost than the equivalent AWS bill. For me, that's too broad a statement, as I have seen it go both ways. What is true: There is only some knowledge overlap between maintaining an AWS stack and having your own Prometheus logged, ceph backed set of boxes.

That is not the case with LLMs. At least, not right now. They roughly work the same and are easy to pick up. They are about as straightforward of an interface as it gets, and using them in "advanced" ways could be summarized on an index card. They are relatively fungible.

I don't see a world where OpenAI runs on brand recognition alone. It needs to be more convenient to run than local LLMs. They've done that by buying so much of the worlds hardware that it becomes more expensive to run these things locally.

I have the same impression. Strange to see this being downvoted & it was after reading the comment that I read the username to find out its antirez!

Now, I think that with these companies IPO'ing and Nasdaq and other bending themseleves and their rules to cater to them (as in case of SpaceX), these companies are very close to an IPO.

So for the employees, they are probably gonna get good evaluations, atleast in the short term and perhaps they are having a problem which is worth having.

But as you have suggested, I feel like the whole thing might be flaky especially given open source models. I believe that OSS models are at worst close to literal SOTA ~6 months ago.

So OpenAI & Anthropic have to somehow always be on the edge to get better models to not lose this (imo) very small time grip that they have, all while losing billions of dollars and having to worry about profitability & so many other concerns in it of itself.

I don't think that there is any other thing inside CS or any industry where two pieces of software being almost comparable enough with not much moat around except a diff of 6 months best, is something on which trillions of dollars float around on. We don't know how things will pan out but if I have to guess, It might not be looking good for OAI, Anthropic over especially the longer horizon.

this is like saying the car with the better engine wins, but all we're doing is commuting to work

Comparisons like that give the impression of reasoning about things, but it's a weak tool to understand reality of very different things.

not really "best" at one thing doesn't mean you win the market for it, rarely does in fact

OpenAI and Anthropic have the know-how for building much larger models that will be a lot smarter and run on datacenter-scale compute. This is a natural 'moat' that will be inherently hard to replicate for on-prem compute or small neoclouds running open-weight/local AI. They can easily coexist with a robust local AI scene.

> if in the future open weight models will be very near the quality of frontier labs, Anthropic and OpenAI will be out of business very soon

> Why would a business pay for Slack when IRC exists?

> Why would a business pay for Dropbox when FTP exists?

AI is not a product per se, it is a technology you can decline into a product, and the product has a lot less value than the technology itself. Who has the best LLM can copy any product idea and make it a lot better. Similarly if open weight LLMs are everywhere and powerful, open source products in the space of agents are too simple to replicate for people to pay big money to a few companies: not everything is alike, not every parallel makes sense. The pi agent is good as a replacement for Codex and Claude Code if you wire frontier models to it. And when products are complex and matter a lot, like complicated AI-powered design suites for instance, there is no reason why OpenAI / Anthropic will win this space instead of a random startup. So either a few companies retain frontier AI, or those companies will die.

About IRC / Slack: other than the fact IRC was abandoned, Slack is about control, not product. The product is terrible.

FTP / Dropbox: this comparison does not make sense.

Bernie Madoff would be jealous. Stealing all open source and reselling "git clone" + "sed" for $1 trillion is something he did not achieve.

The chutzpah is remarkable.

They are selling shovels, not mining gold themselves though.

So it's more like selling a derivative on a promise to steal open source for you in a useful way.

I dunno, the latest Opus models seems to be tuned to waste money... and Claude is kinda lazy lately?

They are far far better at marketing than OpenAI

They just have no moral issues with spamming the internet with bots. They utilize blackhat tactics whenever they can to get an upper hand. Every social media platform is absolutely choc full of Anthropic and Claude promoting bots, and you know they're bots because they all repeat the same things, in the same wording. X in particular seems to have millions of them.

This is an interesting claim to make. Up until quite recently (I mean that in the usual sense of the word, not the AI world sense of the word), almost no one had heard of Anthropic or Claude, despite an reasonably aggressive ad campaign, even at the point when most people would have known about ChatGPT.

Even now, I would guess that if you ask a normie off the street, they are far, far more likely to have heard of ChatGPT than Claude. Of course, Anthropic has been targeting businesses quite a bit harder than the general public for a while, so maybe that's not a fair test.

Anthropic inarguably does make an attempt at marketing their product. But I'm not convinced that the closing of the gap between them and OpenAI (as others have pointed out: I'm not sure it's defensible to claim that either is significantly ahead of the other given the paucity of available data, but they are certainly much closer than they were a year or two ago, when OpenAI was clearly in the lead) is mostly down to that. I think that, for a decent chunk of time (this one I mean in the AI world sense of the term), they had a very non-trivial lead in coding abilities. The developer and business world figured this out and jumped on board. That gap is largely now erased, but that's not enough to retake the momentum.

The models aside, my impression is that Anthropic is winning in large part because of very pragmatic and high-velocity product development on top of them; like with Claude Code.

Like actually iterating hard to make them useful. Many, many details matter here.

I haven't tested the similar OpenAI/Google tools in detail lately though. Previously I found them way too generic and unpolished to be useful.

Is there something to this?

My impression as well. OpenAI was riding the high of ChatGPT with a very confusing and seemingly unfocused offering beyond that. Anthropic was always laser focused on business use cases. Claude Code being the big one. Finance seems to be their next target.

Anthropic has much narrower capabilities. No image generation, no video generation, no 3d world models, barely any voice stuff. But they know who their target customers are, and their API has a model selection anyone can understand and pricing that rarely changes. Focus and predictably

I recently canceled my Claude subscription. The token costs are outrageous. I maxed my 5 hour limit with a single request. If they fix that I'll reconsider but as things sit today Claude is a pig that's just not worth the cost.

My impression too.

Claude Desktop, Cowork, Code, Design all get meaningful new features week over week.

I can’t recall another vendor with such focus and velocity.

Google products are evolving at a glacial pace. OpenAI isn’t as focused on what knowledge workers need.

Velocity: The closest is perhaps NCSA Mosaic/Netscape in 1993-1995. It's exhilarating to follow.

Both Google and OpenAI appear to be stuck in some abstract strategy where they keep shipping new demos instead of iterating on actual products.

Having used both Anthropic and OpenAI models at $work via copilot extensively, I have to say GPT 5.5 currently is best at getting work done with minimal mistakes. However, Claude Code is way ahead of OpenAI Codex in terms of harness features and tooling. MCPs, skills, sub agents, these all were pioneered in Claude Code first. Perhaps that contributed to Anthropic's success.

[dupe]

Main discussion:

Anthropic raises $65B in Series H funding at $965B post-money valuation

https://news.ycombinator.com/item?id=48313048

Other submitted more-common source reports on this from 2 days ago that didn't need traction because of the above discussion:

https://www.nytimes.com/2026/05/28/technology/anthropic-tops... (https://news.ycombinator.com/item?id=48315537)

https://www.theguardian.com/technology/2026/may/28/anthropic... (https://news.ycombinator.com/item?id=48321498)

https://www.wsj.com/tech/ai/anthropic-valuation-openai-80bf2... (https://news.ycombinator.com/item?id=48315537)

https://www.businessinsider.com/anthropic-surpasses-openai-w... (https://news.ycombinator.com/item?id=48316994)

I get the feeling this also means AI works very well for the general coding tasks and that's their biggest success in terms of difficulty AND people paying for it.

Of course every AI company has been over promising and pumping the numbers as much as possible but OpenAI has been hitting the reality wall more because both their people not being able to keep improving at a faster rate and their whole cost structure and financial plates spinning.

This doesn't invalidate the fact Anthropic is also overhyped to the max for their IPO.

Bummer, they are the least friendly to open source, and the most incompatible with free use of your subscription via your own tools/custom harnesses.

Theoretically, I could sell 1 out of 100 trillion shares of my private startup for $1 and surpass all companies on Earth combined by implied valuation. I see people taking the article's comparative framing between OpenAI and Anthropic for granted, but without knowing the private deal terms, all you can really infer is that their 'true' valuations could very plausibly be in the same ballpark.

Except they raised $65 billion dollars, not $1.

Ai is overstated in my opinion but to hand wave the reality of them having created something that investors were happy to value at $1T is pretty unfair

I was attacking the comparative framing of the article. Although there's a lot of private terms we don't know, the claim is taken for granted and overfocused on, either with people saying 'figures out since Altman=devil' or 'no way OAI single valuable company'. I end the comment stating that their valuation is plausibly in the same ballpark.

I've seen fewer people insisting OpenAI has a moat lately, but I'm still not sold the big winner will be either of these two in the end.

I guess the competition lost lots of time in focusing on image and video generation. While they're fun gimmicks, I still really haven't seen the value in AI-generated image/video, especially when considering the greater costs involved.

Doubling down on coding was just infinitely smarter. Has there actually been a successful company which uses AI images and video effectively?

qazinform.com seems to be shadow-banned (and posted only by OP): https://news.ycombinator.com/from?site=qazinform.com

UPD https://en.wikipedia.org/wiki/Kazinform

The headline is false. First off, OpenAI hasn’t raised a recent round so you can’t compare these two companies randomly like this. Second, Anthropic is known to have accounting methods that give it more revenue that they would have if they used the same practices as OpenAI. And neither of these companies are known to be doing gaap accounting.

How much dilution? Who’s getting the value?

I use both, many times at the same time, on different projects.

One day one feels better than the other. Then, by the end of the day, the other feels better than the first. I have no idea why.

I still don’t have a favorite.

In the end, I think both are incredibly useful when I take the time to instruct them properly.

The problems come when I let them run wild.

A sign of a bad developer is they cannot fathom switching from claude cli. In their mind it's claude or nothing. Despite things like codex existings. Opencode existing. Cursor + Compose 2.5 existing.

These are the new .net developers who will know nothing but c# for 20 years.

This suggests that developers are the primary user base affecting valuation, not the average user, doesn't it? I don't know anyone among mortals who uses Claude. The spike does correlate with the exodus from OpenAI earlier in the year though.

All overvalued.

By an order of magnitude.

tesla has been overvalued for almost a decade with little sign of slowing down, it really doesn't seem to matter anymore

It’s because the programming works.

OpenAI. Spent its resources on AGI whilst Claude worked on making programming work.

Google Gemini is out of the race entirely its programming AI is a joke.

It is unclear which strategy will work in the end. 3.5 flash uses fewer tokens and is cheaper.

Gemini is probably one of the better AI's for programming, despite Google not making a decent official client to access it.

I use Claude for coding because I heard it's better, but honestly I can't tell why. It just... works? Maybe that's enough.

Most people think the current valuation is for the models themselves. Actually, they're building the infrastructure for the next 50 years.

The dark fiber of our time?

> Actually, they're building the infrastructure for the next 50 years.

What infrastructure? The hardware would be outdated in 3 - 5 years, after all. What other infrastructure is needed for AI?

Pointless article (like much of the AI marketing hotness and spin room).

> The new valuation is nearly three times higher than the company’s February valuation, when Anthropic was estimated to be worth around $380 billion.

> In March, OpenAI was valued at $852 billion following a record $122 billion funding round.

Basically, today (Late May) we're declaring Anthropic the most valuable. They've nearly tripled in value since February. But also, OpenAI was $852B in March and presumably has grown since then.

In a few weeks we'll either have a new rounding of funding for OpenAI or they'll announce their IPO and the hype train will be abuzz that they're now the most valuable.

my take it's because of the naming: Amodei, Claude and Mythos have this money-throwing vibe to it

Either they are getting fleeced or they are getting very good terms for the investments

Must be ridiculously easy for Anthropic to fill a round, even at that valuation

Some says the founder worked at Baidu before. Is that true?

The core product of AI is labor displacement and wage suppression. Why? To further concentrate wealth. Who are the most expensive employees? Software engineers have to be up there. So Anthropic have been smart here by focusing on a market with the highest potential value.

What I find fascinating is how many inside the bubble defend this for no other reason than they think they're personally going to make their bag out of AI. You're not Sam Altman. Or Elon Musk. Or Jeff Bezos. And you're not going to be.

What's going to happen is that in a few years the sky-high AI salaries are also going to disappear. More work will be done by fewer people in this space too. And only then will many people change their tune because the rising waters have finally reached them.

How long before Anthropic buys Google ?

They're short a few trillion.

This is depressing. Anthropic really is the last company we want to see leading this race, given how greedy they are. Let's not forget all of the lying and gaslighting too. The creator of OpenClaw made this I believe: https://clawd.rip

Stealing peoples tokens because you use a product they don't like... That shows the morals they have. Actions speak louder than words. Disabling peoples caches because they disable telemetry was another juicy one that I don't believe is on this site. In fact there are far more I remember that aren't even listed here.

From what I understand, the Pro and Max subscriptions are kind of "subsidized"; you get way more tokens than they would normally cost. But I think Anthropic wants people to use that for Claude Code only, not for things like OpenClaw. You're still free to use OpenClaw through their API subscriptions.

> The creator of OpenClaw made this I believe: https://clawd.rip

FWIW, it's from an AI chatbot company.

https://twitter.com/theo/status/2048628814755012945

The bubble keeps getting bigger!

T is the new B.

I think claude is much better than Chatgpt

This is all dumb, it's like picking your fave cryptocurrency

Investors of both should read this: https://open.substack.com/pub/sublius/p/srt-introspect-why-c...

"Investors who have poured hundreds of billions into closed-source labs are betting on an unprovable safety moat".

Nobody is investing in closed-source labs for safety reasons, being able to explore more in details what and how the model is thinking is nice but by no means a game changer. What matters to investors and most of the users is that the model gives the right answer at the end.

they don't care, they're driving towards a cliff full speed and are all counting on jumping out at the right moment

Yeah about that

Start what?

I think claude is much better than chatgpt

ChatGPT dropped the ball for a while that most devs and technical people went to Claude for a year or more, they still probably have the most normie market share + are at least trying to win back some of that delay in their latest model so it'd be interesting to see

The "normie" market doesn't pay for enterprise features though. They might cost more in inference then they make back from advertising.

The real problem is this: no cheap model right now produces a genuinely beautiful, usable UI when it comes to website building. Not one.

And here’s the core tension. The models keep getting better. GPT 5.5 improved. But it also got more expensive. Opus 4.7 to 4.8 has become outrageously priced too, up 50%, and 4.6 was already brutally expensive to begin with. API pricing is a real pain.

What’s missing is any meaningful supply of affordable, democratically priced models you can actually embed into your own service. For me that’s playcode.io, whether it’s the website builder or the app builder. The moment we give users access to these models, the cost becomes a serious blocker. There’s no way around it.

The same dynamic explains Cursor. Why did they go build their own Composer 2.5 model? Because relying on third-party models is simply too expensive for users unless they’re carrying a Claude Code or Codex subscription. So Cursor had to roll their own. It’s a real mess, honestly.

And Chinese models don’t close the gap either. They’ve improved, the free-tier ones especially, which is great to see. But the limitations are significant:

• No multimodality. They don’t accept image input. • You can’t attach a screenshot, show a UI, or hand it a PDF. • They feel heavily stripped down overall. • They’re just not polished. Not even close.

Opus, by contrast, feels like a finished, deeply refined product. Everything else is still rough around the edges. And that’s exactly why Anthropic can charge what they charge: because they actually deliver. That’s the whole problem in a sentence.

I don't see a problem as long as Chinese companies are not banned; expensive providers like Anthropic will prove that value is being provided, encouraging competition, which will take care of problems.

Reply on news.ycombinator.com