Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- Nice research and structuring into 4-tier layer. For providers like Anthropic and OpenAI, subscription is the entry point for all these, right? Besides the measures proposed in the article, can token usage % determine these clusters of accounts?by bg24
- There are probably various metrics like language used to prompt the model, number of hours per day spent prompting, and many others.
There are quite a few other mitigations that could be done by providers that aren't mentioned in the article.
by jfim - Ah I was wondering how a certain chinese site I came across did this and it could be this.
That site offers substantial free tokens, is often reported as being flaky and their affiliate links are popping up on different social medias but look sketchy as anything.
by hahahaa - This is pretty fascinating!
Here are the two open source proxies listed in the article: https://github.com/songquanpeng/one-api and https://github.com/QuantumNous/new-api
by simonw - > For example, one operator’s price-comparison site listed a package that bought the equivalent of $3,333 worth of official Anthropic credit for 425 RMB — roughly $0.13 of usage per $1 spent.
Do these numbers make sense? $0.13 usage per $1 spent?
by robluxus - Yeah, I should probably clean this up. The sentence is a bit hard to understand. What I was trying to show was the steep discounts offered by resellers.by mlenhard
- If you are using a stolen credit card to buy tokens and resell them, then the cost per token is the amount the credit card cost you (and building/running the proxy service), not the value of the tokens themselves.by latchkey
- Also 425 RMB is about $59 so $1 of tokens for $0.017 not $0.13 (the discount rate quoted also seems off).by mmoskal
- I think this was very poorly worded. I believe they’re trying to say you pay the reseller $0.13 to get what costs $1 at the upstream provider.by zaltekk
- How do the users know they're getting what they're paying for?
I disabled automatic downgrading/rerouting because it sometimes takes me a second to tell when the answer came from a different model than I wanted. You could easily sell Opus as Fable for a good while.
by blfr - i doubt the users really care as long as they get the job done, AI in 90% of use cases is just a race to the bottom, nobody cares as long as it's cheap and gets the job done.
- You can't, it's all reputation based. Similar to whether drug users don't really know what they got were diluted or not.by gruez
- Seems to be an LLM megaexpansion of the actual source (in chinese): https://www.v2ex.com/t/1196011by nojs
- the way i see it there are 3 types of resellers. the ones using fake credit cards to rack up costs and then cancel the card are doing actual fraud. then you got mass free trial abuse which is more of a gray area and i would say its still wrong. but if you sign up for a subscription, pay for it and resell your monthly tokens thats not at all unethical, even if its breaking their terms and costing the provider money.
imagine ford starts renting out company cars at a huge discount so they can get people to buy the same model for themselves after they drive it at work. its the exact same car and costs the same amount to make, they just take a loss on it and use by anyone other than employees is banned in the contract.
some small company realizes they dont really use their cars that much so they rent them out again for 3 days a week to get some extra cash. is that fraud? it costs ford nothing because they get the same payments either way, they just lose potential profits. they are the ones who decided to set up a loss leader and take the risk of someone "abusing" the system so we dont need to use public resources to defend their strategy. that wastes taxpayer money to protect corporate profits, and it creates moral hazard because ford (anthropic) is not the one paying for enforcement.
by tancop - Instead of credit card, prepaid stablecoin purchases would solve the problem. No chargebacks.by miohtama
- >some small company realizes they dont really use their cars that much so they rent them out again for 3 days a week to get some extra cash.
I mean, in a more accurate analogy, Ford would be paying for the petrol too.
Inference is expensive and Anthropic did not agree to provide inference to some random third party so that the subscriber can make a few extra bucks.
by AussieWog93 - Selling your sub tokens is against ToS and somewhat like sharing your gym pass.by hahahaa
- > some small company realizes they dont really use their cars that much so they rent them out again for 3 days a week to get some extra cash. is that fraud?
Most likely, yes.
There's a common fallacy that once you pay someone for a service, you are free to do whatever you want with that service. In the case of the rental car, the contract the company entered into would prohibit reselling the services and limit who can drive them and for what purposes.
Some people see these limitations and scream "Not fair! They paid money, they can do whatever they want!" The misunderstanding is that the price they paid was predicated on the specific use. They got a lower price for the rentals because the provider calculated the expected use case and priced it according to that.
If the small company starts renting out the cars to try to maximize how much they're used, that breaks the financial model. That's why this type of use is forbidden in every basic rental contract.
It's the same reason why you can't rent an apartment building and then turn it into an AirBnB. On a smaller scale, it's why you can't go to an all-you-can-eat buffet and load up on food to carry outside to your 5 hungry friends. This type of pricing is everywhere.
There is a vocal online minority who believe user license agreements shouldn't be enforced and individuals should never be considered accountable for following them, but that doesn't even apply to these resellers. This isn't a lowly individual user trying to get back $10 from their $20 per month plan that was going unused. There's no way to even achieve the scale and discounts without mass, automated fraud. They're doing chargeback fraud or using stolen credit cards.
It's not even a crime where the big corporation is the only victim. The higher the volume of fraud on the subscription accounts, the less real usage you and I get for our dollar. These people are jumping on the accounts targeted to individuals like us and abusing them to sell tokens to big corporations trying to abuse them at scale. People like you and I lose when these accounts get their limits reduced or the companies start introducing ID checks and KYC just to use basic services.
by Aurornis - This is the problem we've been working on solving with WorkOS Radar. We run it for Cursor and a bunch of other AI companies who have a free trial that gives some free inference to test the product.
It turns out to be a pretty complex program to solve at scale. Token fraud is a lucrative market and the adversaries are surprisingly sophisticated. It's a cat-and-mouse game, accelerated with AI.
(If you'd like to work on this, we are hiring :))
by grinich - I don't think device fingerprinting is the right approach here.
Client-side detection can always be sidestepped, and you need to intermediate the actual inference to get enough signals to make an accurate prediction. There are hundreds of listings for cursor tokens/credits right now.
We use canary values to detect the resellers, and I believe that's the only approach that will actually work at scale.
by mlenhard - It's the same fundamental problem as "ticket touting" for popular events - if you sell something that's in demand at a price that's far lower than the clearing price of the market, you're creating a juicy arbitrage opportunity that sooner or later somebody is going to try and exploitby ifwinterco
- Yeah but those tickets are never a loss leader. So it feels different.
This is more like sharing Argentinan $2/m Google Premium subscriptions via a load balancer.
by hahahaa - Not necessarily.
What OpenAI and Anthropic are selling — a flat-rate subscription with both 5-hour and weekly rate limits — is a bit like an all-you-can-eat buffet.
They expect some customers to generate more in costs than they bring in revenue, just like some people at the all-you-can-eat buffet eat more than they pay for, but by the law of large numbers, the mean cost per customer comes out to something the labs are comfortable with.
What the resellers are doing is undermining the labs' assumptions that every person needs to eat and sleep, and hence won't use every 5-hour window to the fullest. It's the equivalent of buing the all-you-can-eat pass for one person, coming into the restaurant with three of the largest suitcases you can find, and filling them to the brim with food, which you later re-sell at much lower prices. In other words, fraud.
by miki123211 - The real problem is subscription models. Businesses want recurring revenue so they try to game the ratio of fixed subscription prices to COGS but it's always a game and so whoever can figure out the upside for the company can figure out the complementary upside for themselves.
How would one even word a bulletproof subscription contract for agentic tokens, anyway? You can't forbid automation because sub-agents are automation. You could forbid "using tokens for the benefit of more than the human who signed up" but then what do families (especially with kids) need to do? What if your friend asks you a question and you turn to a chat model? Forbidding "reselling" tokens outside of a household sounds like the closest terms but that's leaky for anyone who travels a lot, etc.
Fixed cost per token simply works.
by benlivengood - > How would one even word a bulletproof subscription contract for agentic tokens, anyway? You can't forbid automation because sub-agents are automation. You could forbid "using tokens for the benefit of more than the human who signed up" but then what do families (especially with kids) need to do? What if your friend asks you a question and you turn to a chat model? Forbidding "reselling" tokens outside of a household sounds like the closest terms but that's leaky for anyone who travels a lot, etc.
reverse the pricing structure; give modest discount once you go over certain amount of tokens, then you are incentivized NOT to start multiple accounts.
require first few transactions to be pre-paid to get around at least some of the card problems.
Of course, that would fuck over subsidized plans, but I don't see any option to keep them if you want to avoid the flood
- Subscription models are fine if profitable individually. It then is an automated token allocation.
The problem is loss making subscription as a marketing tool. But if you do loss leaders that be the risk you take.
by hahahaa - The abuse is factored in to pricing and quota structure.
I have some past experience with subscription plans for a much less interesting product. Abuse is inevitable. As you do your math on the subscription costs you look at the actual usage across all accounts, which includes the abuse.
Cleaning up abuse was still a priority because it meant we could give more service to the real customers. It's a frustrating battle because you actually want to give good service to the real customers, but you also want to let each account do as they please with their susbcription. That latter priority probably fades fast for something like an LLM company when you discover that the abuse has become automated and is scaling up so fast that it's tilting the math toward degrading service for everyone.
> Fixed cost per token simply works.
As a consumer, I benefit greatly from the subscription rates. There's a lot of grumbling about how they should go to fixed token for everyone but I'm over hear happy with the subscription plan offerings while they last.
by Aurornis - One aspect that seems to be missing is the abuse of the free credits provided for new companies by AWS, Azure, and other providers.
I know of a friend's company in India who purchased inference, at 4% of the actual price and states that it gave him an unbeatable competitive edge in their large running video influence pipelines. Any new competitors could not offer their pricing at all.
Primarily that operated because registering a new company getting free AWS credits was a very tiny cost
by namanyayg - Video influence pipelines from India huh?
No wonder social media is so shit nowadays. All that brainwashing and propaganda from third world countries, now at 4% the price!
by juleiie - the in India wasnt even needed no worriesby 4848488448