Hacker News Reader

Comment by franticgecko3 | Hacker News Reader

Claude Code burning a large amount of tokens for no good reason is a headline we've seen more than a few times now.

Time to start questioning the motivations of model providers shipping their own harnesses.

It's not really a secret, the answer is profit. And in the part of the world where this provider is from, greed is reckoned as a good thing.

Really? Because you'd think for profit they would want to minimize token burn and therefore get away with lower limits on their plans

That only applies to subscriptions. Enterprise customers pay API rates and therefore per-token. Being wasteful would directly increase Anthropic's profit, assuming they're not selling them at a loss.

You meant to write "being frugal", I think.

I meant "being wasteful with tokens", as in, Anthropic makes more profit when Claude Code sends many input tokens and requests many output tokens when billing the customer for the API cost.

Ahh, sorry, my bad.

As long as price elasticity is zero.

Overall I find this line of thinking pretty dubious. Competition exists. Yes, model provider X can run up my bill by instructing its model to do busywork. But I can easily swap to providers Y, Z, A, B, C.

I happen to find I'm receiving exceptional value from provider X, so I don't switch.

API customers can swap to alternate providers even more easily than subscription customers.

This is not about the bottom line, it is about proving to the marked that there is a demand - in order to attract more funding.

What market is going to care about the number of tokens somebody uses? That's like a market that cares about the number of if statements in a piece of code. It's nonsense.

If Anthropic makes money on each token, the market would/will care very much.

So what you're saying is that they're not actually trying to maximize profit, they're just trying to maximize profit to prove that they can maximize profit for investors?

But they don't. If that was how making money with LLMs worked, everyone would just generate models that generate tokens that do nothing. But of course that is not how it works and they don't.

It's just really bad to do something with a lot of waste that someone else can do with less waste.

I disagree because with their subscriptions and people using Opus I doubt the majority of users are hitting their limits often so this would just burn compute on their end

They need to prove there is a demand in order to attract more funding. Funding they are completely depended on.

Do you really think there’s not enough organic demand so they have to fake it?

I run three 20x max accounts to 100% every week. I could easily use more if I could afford it.

And obviously they wouldn’t IPO if they didn’t need to raise cash, but have you read the S1?

No, surely some feature which is already just using cached input tokens is the secret to their next funding round

Enterprises pay API rates though and they account for the vast majority of revenue

The vast majority of revenue from Claude Code or from their integrating the API in their own apps?

API vs. Subscriptions

I don’t know if there’s any openly available statistics on how API costs divide between Claude Code, Claude SDK, raw API, and via OpenRouter.

All the programmers that work in companies with > 150 employees are forced into API pricing.

You could decide where most work.

I would just say don't underestimate how many consumer/one person companies/freelancers have a Claude sub right now. Especially with all the AI hype on social media its a real frenzy. Would definitely be interesting to see a breakdown. My company spends way more on the API we integrate into our own app compared to our Claude Code sub costs so that's why I was trying to make the distinction across different API usage

Our company doesn't integrate API at all. And we have claude code (obviously with API rates).

Yeah I'd like to see a breakdown because I think there is a huge boom in single person subscriptions that is not sustainable long-term (they will churn high like consumer churn rates, not traditional b2b)

They published the numbers. 83% of revenue is API rates, 17% is subscription (individual and small business).

Lots of end-user API usage is from consumers and small businesses though, like a lot of my business's users

25℅ of all their revenue comes from just 5 companies

https://www.forbes.com/sites/fionariley/2026/10/26/top-surpr...

They are entirely living on enterprise API rates

It leads to the same thing (but probably better spells it out), but the answer is entry barriers to competition which in the long run translates into less competition in the long run and greater profitability

I sometimes wonder how AI labs would make money if the internet didn't exist.

Maybe we'd have local models ...

I don't think today's AI labs would exist without the internet

Then my question is: what would AI labs do if the internet suddenly stopped existing?

They'd have to hire pentesters to run around, inserting pendrives into companies' computers. The computers would get hacked and press releases would brag about how many crimes their models committed. All to prove how dangerous the models are, in order to confirm high market valuation of these companies.

But how would they make money? (They _are_ making money as Anthropic's IPO filing recently showed)

What would software companies do if electricity stopped working?

If the energy grid was destroyed? Give every user a local generator.

The question is: if the internet was destroyed, would AI labs give us local AI?

I don't think it's profit, I think it is genuine functionality. I think that they build features for their internal users while having no token concerns. And this mindset of using LLMs to solve product problems leads to useful features (e.g. agentic auto-approval). Then there is a bit of an issue when they try to ship a genuinely useful feature but now token cost matters.

> I don't think it's profit

Hahaha good one. Profit is the whole reason of existence for for-profit companies. Sure, companies might start with some ideologic grain of thought. But as soon as you reach a certain size, all that goes flying out of the window.

I agree at the top, and it certainly trickles down to some degree. Still, a lot of the everyday engineers who actually make the stuff at every place I’ve ever been just care about making things that are good.

Well, it's either about "profit now" or "profit later". The latter one can feel a lot like it's not about profit.

That is true, but it is definitely about profit. The prior comment made that question via "I don't think it is for profit". Very clearly it is.

You're taking that phrase out of context. If you look at the parent and grandparent comments of the one you originally replied to, it's pretty clear that "the answer is profit" was pointing to the insinuation that Anthropic is deliberately causing Claude Code to waste lots of tokens for no reason other than short-term financial gain, and that's what solidasparagus was disagreeing with.

But would you argue that it is neither about short term nor long term financial gain? Cause that was my point. It really always is about profit one way or another, in my experience, with barely any exception I can think of.

even if their motive is "make our product better", that is still just the profit motive with extra steps. the reason to make your product better is to make it more competitive so that... it makes more profit long term.

Of course it is for profit. Tokens cost money. Just the energy cost alone.

I don't understand this "analysis". Why would it not be about profit?

"Features for internal users" makes no sense here, as it does not apply to the case. "Useful features" is also debatable - LLMs providing user data upstream to Anthropic is a use case for Anthropic, but less so for downstream users and token burners.

Not about profit in the sense that nobody sat down to say "let's add a feature that looks useful and burns more tokens so we make more money". For status update thing I can believe that.

I don’t feel like reality is so interesting that CC devs are sitting in a room theorizing how to extract the most profit from their users.

I’m much more inclined to believe they just don’t _care_ about cost, and profit is an ancillary benefit to their functionality above all else.

Just recently they released auto idle-compaction for large sessions about to miss the cache window. If they were so profit driven I can’t imagine they’d have built this to begin with, that’s free tokens from your perspective.

A smart company would A/B test changes and choose based on lifetime value vs. average session value. The compaction is going to happen anyway, does idle compaction increase the length of a session or the amount of sessions per user?

> I don’t feel like reality is so interesting that CC devs are sitting in a room theorizing how to extract the most profit from their users.

Do you really think this is so interesting, or something that people who work jobs have a responsibility to do?

You’re making a straw man argument out of the nature of what jobs are for, of course they are to help a company make a profit.

I don’t think the CC’s teams have hidden motives to optimize for token inefficiency as the OP I replied to implies.

A reality where a secret cabal of CC engineers are trying to find the most token inefficient ways to build the best harness on the market sounds quite interesting to me, but not realistic.

Somebody in the chain is likely getting profit sharing and, if so, definitely thinks about it (because that’s the whole point of such arrangements).

Is doesn't even make sense, the limiting factor for Anthropic is GPU count, not profitability. They want more adoption at fixed GPU count which demands higher token efficiency.

> Of course it is for profit. Tokens cost money. Just the energy cost alone.

That only counts if people wouldn't find some other way to spend tokens anyway. I regularly have to pace my usage to avoid hitting the weekly limit on the 20x plan; I have plenty of things I could automate.

Jeavon's Paradox means that until all possible uses are saturated, the cheaper frontier AI becomes, the more of it they can sell.

I highly doubt it's the case that it was purely driven by profit motives. I think it's much more likely that Anthropic engineers treat tokens in a similar manner to how we treat memory and compute.

Older programs were MUCH more memory and compute conscious because memory and compute were actively limiting factors. Anthropic engineers probably don't feel the token efficiency requirements nearly as much as we do because they are working with the equivalent of a super beefy dev machine.

an inteligent summary is a user feature, that is not a matter of opinion

They're still getting their $100-$200, why would they go and intentionally burn more tokens at their own expense?

Quality of life comes first, and like any agentic contribution, performance comes later when people complain enough.

I am the last person that will defend Anthropic, but I don't really think they are malicious.

Claude Code is a shitty harness.

Exactly, using 40k of your context upon startup is bonkers.

What would their motivation be to incur the same level of operational expense while giving users less value, making them more likely to switch to a competing product?

I’m sure there’s some galaxy brain theory for why they’d want to keep costs high and reduce value rather than just having an efficient harness and reducing quotas by this 16%, which would be identical for users and save money.

But a far more likely explanation is: a dumb bug.

Haven't we always been questioning that? (Or at least, I thought lots of folks on HN were). There's always been conflict of interest when providers sell both a commodity and ways to use that commodity in a way that benefits the vendor.

I'm also getting tired of this. But I don't know where to go instead.

What is the best drop-in OSS replacement that has support for skills and mcps?

Pi is probably still the gold standard to build on. Or Hermes if you want to go hog wild with connectors and memory.

Every harness has skills & MCPs.

I think omp.sh (oh-my-pi) is the best by a comfortable margin and its completely batteries included. It will pick up and use your Claude Code skills and plugins by default.

If you like to customize your tools Pi is the best, it has a huge plugin ecosystem (all can be used in OMP as well, but it has opinionated answers for most things already). Stay away from OpenCode, its a pile of slop like Claude Code.

Reply on news.ycombinator.com