[prepend]

> $2,180.16 worth of tokens for $200

“Tokens” don’t have an intrisic cost or value. Saying that I used $2,180.16 worth of tokens is like relying on the salesperson to convince me I’m getting a billion dollars worth of pots and pans for $19.99.

I think it’s funny how we are throwing critical thinking out the window when it comes to evaluating biased sources of info.

[simonw]

I'm not sure what you're pushing back against here.

I spent $200. If I had been paying API pricing it would have been $2,180.16. The article is about how enterprise customers get charged API pricing, which means if I had been employed by one of those companies I would have cost them $2,180.16.

What am I missing?

[eqvinox]

Just because API pricing would've been $2180.16 doesn't mean that's the value of those tokens. For starters, you personally probably wouldn't have paid that. But also, sales price isn't value. This is like saying, oh, I saw this bar of gold somewhere for $10000 but got it here for $1000! So I got $10000 worth of gold for $1000! - no, the value of that gold is determined by its weight, which wasn't even mentioned.

We have no market convergence on tokens yet (and it'll differ between LLMs), so it's impossible to say what value you got for your $200.

[aspenmartin]

He's saying he's getting a great deal...a token from Opus on Claude code is the same as a token from Opus on the API. I remain as confused as Simon. He's not talking about "here's the ROI I got from my $100 subscription" it's "here's how much I saved from getting the monthly subscription instead of sending things through an API".

[mjr00]

Right, the confusion is that the quote-unquote "subsidized" monthly pricing is often used by Anthropic/OpenAI skeptics as proof that inference is unprofitable, i.e. the API would have cost $2000 but you only paid $200 for a subscription, therefore OpenAI is selling dollars for 95 cents and the house of cards is about to collapse. As the GP says, this is faulty logic because we don't know what the actual cost of a token is; OpenAI might only pay $1 in inference costs, in which case they're merely "incredibly profitable" making $199 off you instead of "ludicrously profitable" making $1999 off you had you used the API.

But to your point, re-reading the article, this is not what Simon is saying at all; he's just pointing out that he got to use ~$2000 "worth" of tokens on his $200 plan. Which makes total sense! Subscriptions are sticky, that's why the entire software industry moved towards subscription models (as much as we hate it); the person paying $200/month is more likely to stick around than the person who paid $2000 using the API.

[brazukadev]

Which is a terrible metric sending a wrong message.

[aspenmartin]

He’s just making a point that $100/mo subscription is cheaper than the API costs and that the same work would cost 10x as much on API pricing. It’s not some sort of swindle it’s a trade between consistent revenue that power users will overspend on but plenty of people will underspend (I’m on the underspend side) or pay-per-request service that makes more sense for say scaling out something custom (I run APIs a lot when I have say some pseudo labeling pipeline I want to scale to a lot of data. For this I typically use smaller models but lots of cases where you want larger models). What’s the wrong message here?

[jonnat]

No, value is determined by what participants in a market are willing to pay for something. The only reason you are able to say that the value of gold is determined by its weight is that gold is a commodity and no matter what you paid for it you'll find others willing to pay market price.

Simon is saying that companies are (today) willing to pay API prices for tokens which is as good as any determination of value.

[yokoprime]

Is this some anti-FIAT take? The value the author got is not value as in intrinsic value, it simply means value as in better deal than the alternative. This is often called "value" and you will see this used when products are sold in "value packs" etc.

[remus]

> Just because API pricing would've been $2180.16 doesn't mean that's the value of those tokens.

You seem to be suggesting the price of tokens is entirely disconnected to the cost of providing the service? I don't see much basis for that assumption.

[recursive]

I'm willing to charge you $100k for those same tokens.

Does that mean you'll be saving $99k?

It sounds an awful lot like the mark-up to mark-down scheme where the price stays the same.

[OrangeDelonge]

Large enterprises make deals and won’t be paying 2,180.16$ either. Just like with AWS

[simonw]

That doesn't seem to be the case. From what I've seen enterprise deals get API pricing now. Have you seen evidence that's not true?

[roomey]

Hi Simon, nice article. The parent there may be making the same assumption I am, that large enterprise _never_ pays sticker price.

Also, to just color in the picture here, as I haven't seen it mentioned elsewhere, there is a very large Saas company at the moment who has given everyone unlimited tokens on Claude. And they have a dashboard showing who spends the most. So the "budget" went from about USD500 per per person (split between Claude and cursor) in Jan to... Well a soft limit of USD100k... Per month... Per person.

People can still see the top line sticker price on their spend, but honestly I can't believe that the Saas is paying that full price when the invoice comes in.

That said, there are some finance reports which are probably dropping soon where we will find out!

[simonw]

> The parent there may be making the same assumption I am, that large enterprise _never_ pays sticker price.

I shared that assumption until yesterday, when I found out that it wasn't holding for LLM pricing from OpenAI and Anthropic. That's what inspired me to write this piece.

I think those token leaderboards are an obviously terrible idea and will go extinct very quickly now that people are paying attention to costs.

[wongarsu]

But the feature list at https://claude.com/pricing#team-&-enterprise literally lists "tiered incentives on committed spend" and "non-standard terms" as perks of the sales-assisted Enterprise plan. Maybe "non-standard terms" could mean "we dance for you if you pay", but what would "tiered incentives on committed spend" mean besides "we can negotiate on price if you bring the volume"

[tomjakubowski]

Anthropic advertises volume discounts:

Volume discounts may be available for high-volume users. These are negotiated on a case-by-case basis.

* Standard tiers use the pricing shown in Model pricing

* Enterprise customers can contact sales for custom pricing

And there are discounts available through "Claude Platform on AWS":

Anthropic rates your token usage in USD at standard per-model, per-feature rates, applies any negotiated discount, converts the result to CCUs at $0.01 per CCU, and reports the CCU quantity to AWS Marketplace hourly. Your AWS bill shows a single CCU line item.

https://platform.claude.com/docs/en/about-claude/pricing

On the other hand, contrary to Anthropic's documentation, another source claims they've killed pre-existing API volume discounts for large enterprise customers as of April: https://itbrief.news/story/anthropic-shifts-enterprise-billi...

[asib]

> > The parent there may be making the same assumption I am, that large enterprise _never_ pays sticker price.

> I shared that assumption until yesterday, when I found out that it wasn't holding for LLM pricing from OpenAI and Anthropic.

This reads like GP saying "enterprise never pays sticker price" and you responding "I thought so too until I saw the sticker price".

Is there some info you have that you can't/didn't share? Your article doesn't offer anything beyond the above.

[mvanbaak]

large enterprises dont pay openai or anthropic, they get this thing called copilot and get a nice price there. At least on this side of the pond (eu)

[mrandish]

> I found out that it wasn't holding for LLM pricing

You're correct. When a type of cloud service grows large enough and has a few competitive suppliers, enterprise pricing tends to coalesce with the large buyers paying around the same price for the same thing. While that might be lower than the publicly cited rate card, the private price similarly large customers pay ends up being similar.

One reason is that the very largest, long-term enterprise customers are so valuable, they can command MFN clauses ensuring no one else is paying substantially less for the same thing, then the rest of the rate card for smaller customers flows down from that. There's a strong disincentive for vendors to cheat or allow big disparities between similar classes of customers because the number of people involved on both sides of these deals is large enough that word will get around eventually.

Large scale enterprise sales and purchasing in a given sector tends to be rather circular. Account execs move to other vendors and call on the same customers, while purchasing agents can move to other customer firms. Personal relationships, reputation and credibility matter. Lying to screw a large customer over just to make one commission or quarterly quota can be a very bad long-term career move. Sometimes purchasing agents or executives quietly compare notes off-the-record with their peers from other similarly-sized firms. After dinner drinks at industry association meetings and trade shows can be quite productive in terms of verbally exchanging 'market insight' with peers.

When there are significant pricing differences, its usually due to different volume commitments, SLA/QoS guarantees, payment terms and other material factors which justify the difference. Source: been there, done that inside a top ten valley tech company. Once was in a meeting where a newly minted EVP tried to get a long-time senior account exec to pressure a huge customer by being semi-dishonest. The account exec schooled the EVP on the fact that the EVP could only make him unemployed for about 8 hours but that huge customer not wanting to work with him could make him unemployed forever. :-)

[themgt]

I do know of moderate-size companies deploying OSS LLMs on their own GPU clusters, for ownership/security/maybe cost reasons. I'm somewhat surprised F500 companies are apparently just handing over all their data to the model providers.

Could be fantastic for small shops while it lasts. The big guys have to pay 10x for precious tokens.

[no-name-here]

I guess F500 companies’ AI contracts include data protections.

[waisbrot]

And "large" just means that AWS will assign an account manager to talk with you. I was at a start-up who spent $300k/year on AWS and that was enough to get special attention and discounts. Enterprise pricing is confusing.

[apsurd]

The point is that those a real prices real people are paying for real API usage. it's not made up.

your point is large players won't pay those prices at massive volume. ok

[Anon1096]

Claude is so in demand at the moment that there aren't really volume discounts. Anthropic sets the terms and you either accept them or get lost they have that much of a lead (mindshare/desirability wise).

[yokoprime]

They pay sticker price. There may be exceptions for very very large companies like Amazon or Microsoft which have their own deals where they rent out compute in return for usage.

[pembrook]

API pricing drops DRAMATICALLY in enterprise agreements.

As with pretty much anything priced on volume/usage.

Enterprise deals are negotiated ad-hoc, the listed pricing is simply a jumping off point for the final negotiated discount.

If you’re going to give 20,000 employees Claude code you are not going to be spending $1B per year on Anthropic tokens as if you gave everyone an individual API key. Just as Anthropic isn’t paying AWS SES $10,000,000 to send 1 email update to their massive user base when the next Claude version drops.

[taude]

This isn't true at the moment, though. So far there hasn't been the negotiating power. What happens is you end up capping usage for employees at a fixed amount. I think eventually, prices will come down and there will be discounts, but for enterprise accounts at least of our size (<5000), we're paying almost 100% retail, which kind of sucks, because it's expensive, and pretty easy to burn $50 to $100+ in a day, if you're not careful. In fact we got pushed off the former plan to the token-utility one at the last contract negotiation.

Going to be interesting to determing the metrics we give to engineers for determining whether the spend on this is worth it. Measuring PRs, lines of code committed, commits fully generated by agentic workflows, etc.....

[yokoprime]

As someone who has seen the enterprise deals, they are not subsidised in any way shape or form. Anthropic may wave the seat fee, maybe get lower "expected" consumption. This changes what the company pays up front. but token prizing is fixed.

[tomjakubowski]

To confirm, you're saying the enterprise deals you've seen match the prices per token listed here?

https://platform.claude.com/docs/en/about-claude/pricing

[simonw]

> API pricing drops DRAMATICALLY in enterprise agreements

Do you have any numbers or reports to back that up?

[lrae]

> Just as Anthropic isn’t paying AWS SES $10,000,000 to send 1 email update

How much do you think emails cost? That number is just so far off?

But besides that, running SES is also quite a bit cheaper than SOTA ai models with high demand (and comparatively) no competition. And quite a bit more pressure to make money (soon).

[tomjakubowski]

I think it was a figurative example. For what it's worth, $10,000,000 buys you 100 billion (1e11) outbound emails on SES at the sticker price ($0.10/1000 emails). One source puts the number of emails sent worldwide in 2024 at 132 trillion (1.32e14).

https://aws.amazon.com/ses/pricing/

https://www.statista.com/statistics/456500/daily-number-of-e...

[no-name-here]

But if their “figurative example” was the opposite of reality, then we should similarly assume their other claims were the opposite of true as well?

[tomjakubowski]

How is it the opposite of reality? The commenter exaggerated some numbers, or charitably made a math error, but as far as I know the thrust is correct: AWS is well known for offering discounted rates to high volume customers. In some parlance I hate but will adopt here: the comment is directionally correct.

Do you have any evidence that AWS doesn't offer volume discounts? I have seen plenty of evidence that they do.

Speaking to the subject of this sub-thread, there is pretty good evidence (Anthropic's own pricing page) that Anthropic also offers volume discounts on their API pricing. Five days hence, they have not taken down this language from their platform/API pricing page:

> For higher rate limits or custom pricing arrangements, contact the sales team.

> Volume discounts may be available for high-volume users. These are negotiated on a case-by-case basis.

> Enterprise customers can contact sales for custom pricing.

Simon has ignored all of my comments arguing this point. That makes me reconsider his credibility: why won't he provide any evidence to support his assertion, or admit he may be wrong here?

[altruios]

> If I had been paying API pricing it would have been $2,180.16

The point being made above is that API pricing is calculated... somehow... seemingly arbitrarily. Possibly untethered to the infrastructure costs entirely: which would be the basis of any 'value', however that holds the labor theory of value, which isn't accurate either. So how do you accurately price these tokens at all (other than through price-discovery: which is slow, messy and fuzzy)?

[NitpickLawyer]

> So how do you accurately price these tokens at all

Like anything else in the economy: at the point where enough customers can pay you, and not enough will go to the cheaper competition.

[altruios]

> at the point where enough customers can pay you

> (other than through price-discovery: which is slow, messy and fuzzy)

I notice a distinct lack of reading or comprehension (from everyone around me now, not just this comment) which worries me. I worry if LLM's are to blame. No one reads anymore...

[no-name-here]

I imagine a number of Hacker News members might be devs that haven't dealt with terms like “price discovery” before, so we should all try to show grace.

[altruios]

My apologies. I was taught that term in 7th grade econ. I assumed it was common vernacular.

[lbreakjai]

I don't really understand the holdup about that specific line? Anthropic publishes the cost per million of tokens, what's arbitrary about looking at your consumption, and doing the math?

Or are you saying that Anthropic is determining that cost per million arbitrarily?

If so, it'd be like asking to explain why things fall on the ground other than through gravity. Companies pick the price they think the market will bear, and adjust based on new information.

[prepend]

Lets say McDonalds charges $2000 for a BigMac. If they offer a deal and sell it to you for $200, did you save $1800?

Maybe if you spend $2000 on a BigMac. But it’s unlikely you would buy such a burger.

What is a hamburger worth? Don’t look to McDonalds to set the value.

[xnorswap]

Have you or I misunderstood the "teams" plan?

edit: I missed the "enterprise" feature matrix with the usual audit/compliance stuff to force the biggest enterprise customers onto enterprise plans. Otherwise the "teams" plan is much better value for any business.

orig-continued:

https://claude.com/pricing/team

Teams premium is "Everything in standard, plus more usage*"

And from my experience, it's a very generous usage, I've only hit the limits once or twice, and both times required multi-boxing agents.

I could single-window agentic development all day on opus-4.7 auto-mode without hitting limits.

If you're a business using claude, then that seems like the right plan, the enteprise/API plan seems more suited to where your product is built on top of the agent themselves, so seats/limits aren't really meaningful?

[nr378]

Claude Teams and Claude Enterprise are 2 distinct plans. Simon is right that Enterprise seats have no included usage (and so all usage is charged at API billing rates), whereas Teams seats do.

[troyastorino]

Tokens do have a clearly calculable intrinsic cost. There's the marginal cost of production (i.e. the inference cost) and the amortized R&D cost that goes into the model producing them.

Yes, value is hard to calculate, but luckily market pricing mechanisms exist exactly for this purpose. There isn't a better number to use than what people are willing to pay for them.

So he's saying that on an enterprise plan, he'd be spending $2,180.16. He's not paying that much, but enterprises are.

[m3kw9]

Enterprises also use 200$ plans, they are not that stupid. They add a spill over API key for over usage.

[simonw]

OpenAI and Anthropic won't let them any more. Once you get above a certain size (I believe 150 seats) they push you onto the enterprise plans.

I don't believe you have the option to keep with the $200/month flat rate subscriptions any more. I'd be happy to be convinced otherwise.

(I dug into this a bit more and couldn't find anything in their consumer terms that say "you cannot use this personal account if your company has more than X people", so I imagine the pressure is more that your big company's purchasing department really doesn't like managing hundreds of individual subscriptions as opposed to a single, stable, predictable negotiated contract.)

[klausa]

The purchasing department _really_ doesn't like it, but it's the corpsec department that will really want to murder you for using "personal" accounts to interact with corporate codebases.

[prepend]

There’s a cost, certainly. I expect Anthropic knows. But we don’t know what that is.

[woah]

From my back of the envelope analysis for my own projects, paying per token on OpenRouter is competitive if not cheaper than running the same open weight model on a rented GPU. Per-token pricing is in the same ballpark (although more expensive) for closed frontier models and open weight models (cents to dollars per million). To me this says that the pricing is somewhat grounded in reality.

[rohansood15]

Are you comparing single-user requests or multiple concurrent requests when you say comparable to rented GPU? Most of the cost efficiencies kick in with concurrent/batch requests. A single H100 node can provide like 5k input + 2k output tok/s on a model like Qwen 3.6 35B-A3B with 30+ concurrent requests.

[rldjbpin]

> “Tokens” don’t have an intrisic cost or value.

i am pretty sure these services know what it truly costs them to serve you tokens, maybe not in realtime but at least periodically.

however, what they charge us is a constant exercise in price discovery. i agree with this sentiment in the sense that we don't have a stable sense of the cost. all of these comparisons are good for the moment, or at most the near future.

i believe that even the "all you can eat" approach with the max plans, regardless of their crazy pricing, is not sustainable only with the power users. if most of us gets this kind of value through our plans, surely it does not incentivise the service providers to continue pushing it. maybe they can regardless just to gain market share, but not forever.

[seattle_spring]

Reminds me of a car dealership talking about how good of a deal their extended maintenance plan and warranties are.

[john_strinlai]

a little critical thinking led me to read that sentence as $2180 worth of tokens [at current api pricing]

[FergusArgyll]

I think it's funnier that you can believe some things have an intrinsic cost and others don't

[dnnddidiej]

His point is more he was surprised enterprises weren't getting the discount. And so indeed maybe it is not a giant ponzi after all! (Could be a bubble)

[jfrbfbreudh]

Lol. They obviously have intrinsic cost, the floor being the cost of electricity. It’s hilarious how we are throwing critical thinking out the window when it comes to evaluating biased sources of info.