Hacker News Reader

Comment by echelon | Hacker News Reader

It it wasn't for competing models, this would all make sense.

If this was just OpenAI and Anthropic, this money would represent all of your salaries being soaked up and your job functions being privately consumed by two companies.

Fire three quarters of your eng staff and subscribe to Anthropic is a valid trillion dollar business. Scary to say, but that's the value prop on the filing.

Except that there's actually tons of competition and no moat. The math problems are just a distraction from that.

How on earth are you worth a fraction of that when there are a dozen Chinese companies doing the exact same, with half of them just giving away the goods for free?

View on HN

It's not the exact same. People aren't giving Anthropic billions because they are idiots or dislike the Chinese.

I don't disagree that Anthropic has no long term moat but they have been a LOT better than the Chinese models since the beginning.

As programmers, we're natural optimizers. And we frequently deliver solutions that aren't always the best, but just "good enough".

Guess what, the Chinese models, Mistral, Argon etc, aren't "good enough" for a lot of tasks. That's why people pay for Claude.

The delta between "good enough" and "great" is shrinking.

As programmers we don’t make the decisions. If anthropic allows the c suite to fire 1000 people but China only 900 people, then anthropic can only coat 10% more than China to be worthwhile.

It isn’t just the Chinese models, though. Google Carbon just came out and from what I have heard is very competitive in direct comparisons and cheaper. Then there is Grok, Muse, etc. I feel fairly confident that there is not enough room for all of these to be successful businesses. Maybe enough room for two at the current valuations, but Google can just lurk in the background forever. Open AI and Anthropic don’t have that luxury. The minute they falter, for even a short period it is pretty much over.

Carbon isn't out yet.

And when it is it will still get stuck in loops.

Argon is definitely competitive (used it over the past week or so).

It's pretty slow, but comprehensive. Caught a bunch of stuff that GLM-5.3 flash (which is incredible value) missed, and gave me good solutions for some irritating issues.

It's very slow (even by frontier standards), but it's hard for me to believe that Anthropic have much of a moat.

I mean, I like talking to Claude more than basically every other model (the open models are much more entertaining though, especially their thinking) but that doesn't feel like it warrants their valuation.

If they can successfully move towards owning their compute, and optimise it massively then they might do well, but it's a risky proposition from my perspective.

I've been using Argon to write documentation, it does laps around Claude which is great for code, but should go to the school for LLMs who can't write good but wanna learn to do other stuff good too.

80/20 - 20 pct of providers serve 80 pct of market. It works everywhere and everytime.

Can you share any example prompts where Chinese models aren't good enough?

I've tried Chinese models extensively over the last 1 year I've been working full-time on this project [1]. They're borderline useless. 90%+ of my code is written by Codex, and perhaps <10% by Claude and myself.

> Can you share any example prompts where Chinese models aren't good enough?

Nobody is prompting any more, it's more planning, automation, guardrails and reviews. For example, if your tests report that there's a 40% slow down in a particular type of analysis, you have to review possible causes, and approve a plan involving multiple agents to find solutions that won't break 50k tests. And that solution itself is a small part of the work that gets planned in a day.

Codex and Claude are the only tools that work for me, inspite of testing almost everything out there to save costs.

[1]: https://github.com/orgs/tsoniclang/repositories

You can't use LLMs without prompting them. But, ok, you use codex and it doesnt feel like you prompt them much, but do you have an example commit from codex which you are sure can't be matched by Chinese models?

Hard to explain this without taking up space on an unrelated thread.

But I'll give you a short example. Let's say you have const x = [1, 2] and we want a policy/solver for the type of x (int32 array, int64 array, or floats etc). We could say we'll infer based on the usage of "x", but it's not free. For example, what if it's passed into a generic function add<T>(...), or if there's reassignment or it's aliased, or if it's being used in a loop counter, or a math operation or a combination of these? This is of course doable, but it requires a sweep of a million lines of code to understand the implications on C#, Rust, Mojo targets, and on the current code and tests.

I haven't tried this specific example with OS models, but this is usually the kind of thing that almost immediately falls short and breaks everything. Codex and Claude will do a decent job of analyzing these and giving me better impact analysis. Codex and Claude will also be wrong many times, but the output is often usable.

Add: When I see OS models coming very close to frontier models in coding benchmarks, I just assume based on my experience that they've been benchmaxxed.

Happened to me yesterday. Create this spec targeting this language and version. It creates it using a higher version.

It isn’t just winner takes all anymore(hasn’t been for years?)… I don’t think any reasonable consumer/company would hitch their wagon to one AI provider unless they came with an order of magnitude better quality at competitive costs.. For example, we use Claude Opus 5.5 for particular tasks that justifies its cost, but use self-hosted(not really… hosted on AWS Bedrock, but the point applies in spirit) cheaper models from Anthropic’s competitors and even open weight models for specialized tasks that don’t need the frontier model. Even if Anthropic pulls a huge lead against their competitors, there is no reality where the whole world sticks to Anthropic and shuns cheaper models. This is why their absurd valuation sucks. At reasonable valuation, they are quite a bet because Opus 5.5 is still a great workhorse as of today.

> People aren't giving Anthropic billions because they are idiots or dislike the Chinese.

Maybe this describes Anthropic people mostly

I would not assume this represents their investors

Kevin O'Leary has specifically mentioned kicking China's ass wrt AI race

There is likely a ton of xenophobia and racism motivating a lot of stupid investment within the US

I'm sure a lot of people tried 1 model/harness and stuck with it without really comparing (same way I haven't changed my IDE and my terminal for the past 10 years), and eventually will try a Chinese model and be blown away.

I was discussing with a coworker yesterday who pays $200/month for cursor, has never tried any other harness and mostly use grok, in rare occasions opus 5.5. He seems to be regularly out of tokens. I told him I exclusively use DS 4.1 flash with opencode, spend like 100-150M tokens a day including cache and it oneshots all my tasks without issues. I barely reach 20% of my $10 sub at the end of the month. From his look I think he didn't believe me.

> People aren't giving Anthropic billions because they are idiots or dislike the Chinese.

There are plenty of people who give Anthropic money because 1) Reddit told them to switch to it or 2) because telling their investors that they use Chinese models (never mind whether they're open-weight or where they're hosted) is a non-starter.

> the Chinese models, Mistral, Argon etc, aren't "good enough"

Yes they are. Earlier this year the gap was narrowing rapidly; it wasn't until this summer that the frontier labs pulled ahead again and them doing so doesn't diminish the value prop of something like the GLM 5 series or DeepSeek.

They really, really aren’t. Opus and Astra are just barely good enough for me a lot of the time.

I was able to get GLM 5.3 Flash to reverse-engineer a 28-year-old Java applet game and re-implement it in Chicken Scheme using SDL bindings (and this was before the current hypewave of whatever Claude RE skill influencers have been peddling). It made some UI and gameplay mistakes that I then corrected with manual testing and a more powerful model but the headline here is that it was able to competently work on a bespoke 3D engine in an obscure language with no trouble. It chews through CRUD work and will happily bootstrap a working MVP given enough steering up front.

Opus 5.5 is capable of the same and then some (it can edit 3D environments if you ask it to). I have no idea what your benchmark is but to say that these models "really, really aren't [...] good enough" is not reflective of my (or, anecdotally, most peoples') experience.

ok, but the valid business prop is to be a dependent on anthropic, indistinguishable from just being a second class citizen, since your entire networth can't be extracted from anthropic.

Go nuts with it, but we're at the SciFi dystopia test with: how do mega corporations even form? Well, this is certainly the plausible reality, they just start aggregating random income streams because their product is essentially the only thing that makes the business profitable. Without it, it can't compete.

Hopefully this sounds skeptical, but I don't see how you can have an independent business and be entirely dependent on third party for your operations.

>> I don't see how you can have an independent business and be entirely dependent on third party for your operations

Like a lumber mill that buys logs from a logging company? Who subcontracts to individual fallers? Who buy their chainsaws from a chainsaw company, their fuel from a petrochemical company? No business isn't entirely dependent on third parties. That's what it means to exist in a healthy economy.

A "farm-to-table" megacorp that owns everything from the source of every raw material through to the finished product is a nightmare, an entity that needs to be reigned in and broken up. Capitalism is a competitive game, but a game that can never be allowed to end.

But what about the digital economy? It's interesting that the length of the supply chain for digital goods and knowledge work is nothing like physical goods. Google, Microsoft, and Apple are end to end, don't have to depend on any other software company if they don't want to (until LLMs, anyway, except Google).

What is a "digital economy"? There is only one economy as far as I know. Division of company inputs into software and non-software is arbitrary and has no merit as far as I can see.

>> don't have to depend on any other software company if they don't want to.

I am rather sure that Google is heavily reliant on Microsoft and Apple to produce the operating systems used on devices that are in turn used to view Google's ads. And Microsoft is heavily reliant on hardware manufacturers (firmware is also software) to make the machines that Windows lives inside. And Apple is reliant upon a thousand and one companies that enable the innumerable mobile networks, which include many layers of non-apple software.

They are not reliant on them second-to-second. If anthropic tells these businesses they can't use anymore, or price hike, or ... then those businesses are kaput, right that second.

Plus you can bet your firstborn everything passes through the NSA first. So seems like a spectacularly bad idea, especially for .eu business.

There's no "right this second" way to kaput Google, not even if you're Microsoft.

The difference is that the lumber mill isn't giving the logging company all the information required to run a lumber mill and the logging company isn't turning around and selling said information to anyone for $200/mo.

In my scenario, the only way anthropic and openai are profitable entities if they directly absorb the capacities of the companies, the same way Amazon "basics" exist to absorb the generic consumer product levels.

I'm not pointing out they'll take your secrets and sell them to the competition. I'm pointing out they will _be_ your competition. The same way Amazon basics competes with it's business interests.

> I don't see how you can have an independent business and be entirely dependent on third party for your operations

This isn't new though. A ton of software companies are already completely dependent on AWS or Azure or whatever.

> I don't see how you can have an independent business and be entirely dependent on third party for your operations.

All businesses are reliant on multiple third parties doing their thing in order for the business to be viable. A lot of companies exist at the whims of Amazon or Walmart.

Not intellectually.

There is no such thing as an "independent company". They are all dependent on the federal government to protect them from the marauding herds.

Except PMCs.

Reply on news.ycombinator.com