Here's what is happening:
Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices. They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs. They are subsidizing model access in exchange for user logs and reasoning traces, which they then sell as training data, allowing them to operate below cost.
Claude and ChatGPT are both blocked in China. You need to use a VPN to access either, and you can't pay with a Chinese bank card. So most people who want access to Claude buy access via a reseller. It's the easiest and cheapest way to access Anthropic models in China.
These resellers operate tens of thousands of bot accounts, which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
Here's one token reseller, they're offering Opus 4.8 at a 93% discount below official API rates: https://yunwu.ai/pricing?provider=Anthropic
This is one reason why DeepSeek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
I shared this story a few months back, but it never got any traction. It explains the token resale economy in China, it's an excellent read https://www.chinatalk.media/p/how-to-buy-cheap-claude-tokens...
>They achieve this by reselling capacity from pooled Claude Max 5x accounts, payments fraud, and also reselling the model output to various Chinese labs.
>Here's one token reseller, they're offering Opus 4.8 for a 93% discount below official API rates: https://yunwu.ai/pricing?keyword=claude
But is it cheaper than getting your own account? Otherwise this sounds like the "anthropic/openai are losing gazillions of dollars because they're selling $1k worth of tokens for $100" line that's commonly trotted out by AI bears.
> Claude and ChatGPT are both blocked in China
So it's presumably cheaper than attempting to spin up your own method of circumventing the blocks.
You can use it as an API unlike the subscription.
It's very difficult for people to create personal Anthropic accounts from China. Anthropic blocks Chinese bank cards, so people must pay with a foreign bank card, which they likely don't have. And even if they manage to set one up, they have to access it via VPN, which eventually gets the account flagged. They then have to complete identity verification, which most Chinese users are unable to pass.
There's a similar Claude resale market going on in Russia. On Funpay they are selling Claude tokens for roughly 20-30x cheaper than official Anthropic API pricing.
And phone number verification too? So that's 3 hurdles to jump to just get opus.
for verification you can buy phone number for $1 easily.
Doesn't ~every phone number verification service check the telephone provider and only allow from a select whitelist of residential providers?
With esim and virtual credit card services, why would that be a blocker? Who's got a landline these days?
You can easily get lots of esims for $5-10 months in EU / UK or just for free.
Phone verification services exist to give you a real number for the purpose of passing this kind of verification.
They even have APIs for these services. They make money since they can use the same number to verify different things.
Maybe these resellers are using stolen American credit card numbers? Reselling Claude access seems to be a nice way to launder the money.
> This is one reason why Deepseek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
This one does not make sense to me at all.
Deepseek and GLM are openweights, even US inference provider are selling them at much cheaper price. The price is cheap because the model is more efficient.
DeepSeek permanently cut its V4-pro API prices by 75% because they were too expensive. Without the price cut, Deepseek V4-pro tokens would have cost more than resold Opus 4.8 tokens.
Opus 4.8 is a more capable model, so almost nobody was going to pay for V4-pro at the original price.
China also have trust issue with American companies. Most of State-owned companies will not use those services even if they can directly access them.
And? The US feds wont allow even local Qwen or Deepseek models either. "Evul godless commies" or some such nonsense.
Urm, no? I man they did cut prices by 75% that part is true - but they reduced a starting price that was below sonnet.
Also it's a open weight model, doing that is impossible long term because the real price will be set by the other model providers, who priced it around 60% of sonnet inference cost. Had to look that up though, so that's today's pricing.
Is there a contradiction here? If resold Opus tokens are sold at a 93% discount, you can be a lot cheaper than Sonnet while also a lot more expensive than resold Opus tokens.
I see, After rereading the comment I was responding to I realized I probably misread/misinterpreted what they wanted to convey.
I think there isn't a contradiction and I was just confused. The price may have been discounted only to get below the price point of opus resellers. I do not have enough information on that to make any clear determination on that topic.
If other providers can match Deepseek's first party prices, that probably means that the economics for running inferencing work out for them.
> Without the price cut, Deepseek V4-pro tokens would have cost more than resold Opus 4.8 tokens.
You mean it's functionally as if American tokens are being price dumped in China and Chinese model providers are being forced to compete with that and innovate? So many delicious layers of irony, lol :-P
If resold Anthropic tokens undercut even the at-cost open-weight model tokens, because they're reselling subsidized subscription tokens, then you'd have to start selling open-weight model tokens at a loss in order to match them.
It's somewhat difficult to have any sympathy for Anthropic here. They're entirely responsible for selling tokens at below cost, with the age-old bait-and-switch tactic.
If they weren't doing so, then these Chinese resellers wouldn't be viable. Radical idea, but how about they actually charge a viable price, even on subscription plans?
Also, wouldn't that claim only hold in China?
I'm an European and I'm not using those proxys the article describes.
One would think Anthropic could point Mythos at this to solve the reseller problem outright:
- Purchase multiple accounts via resellers
- Send messages that contain a UID
- Capture these in Anthropic's logs
- Shut down account. Use any metadata to identify related accounts
/loop
This, just like blanking out a football stream for a split second to binary search and find IPTV rebroadcasters, is far too good a solution. Suits prefer to make it seem like their job of fighting "misuse" is hard, justify their budget, continued existence of the trust & safety department, face scans, etc.
They could be doing this internally and want to see if they can downright eliminate these loopholes before bringing Fable back.
I don't care how they do it, I just want to use Fable again.
Maybe Fable is not as capable as thought?
On the one hand they talk it up as world ending and on the other hand they can't manage bot accounts on their own service.
I want to hear how this can be rationalised.
From the article "every layer of control frontier US AI companies have added (geoblocking, phone verification, credit card requirements, and now live biometric KYC checks) has produced a corresponding layer of evasion infrastructure".
No system is foolproof. They'd have to be willing to throw out some % of good customers along with the bots. Amazon can do that because they have a monopoly already. Anthropic can't risk it when they're trying to grab market share.
In this case, being distilled is sort of existential to them. The false positives would just be losing some revenue (depending if profitable, not even losing profit).
One would believe a model scoring this high on SWEBench could maximize F1 score for a precision recall problem easily. What's the missing part?
This only shuts down the account you have bought in the first, plus a few others if it is shared.
> Use any metadata to identify related accounts
How does that work? I think this is the most important part to have an impact on the „thousand“ bot accounts.
They probably will route you through different accounts. So with a single account you should be able to hit hundreds-thousands of accounts.
Especially if you did this in a sting-like manner and used subtle compound UIDs (read: unique combinations of innocuous words) then banned every account found at a later time.
Rather than go after accounts, you’d go after the hosting facilities. The hosting companies are more likely to be able to identify related accounts so working with them will give more scalable results.
> One would think Anthropic could point Mythos at this to solve the reseller problem outright
You're assuming Anthropic want to stop it.
I think it serves their interests more to be able to release stories like this from time to time, to feed to the US government, in an attempt to get the Chinese competition shut down.
No, it more likely serves their IPO story with growth metrics.
One doesn't ban a large chunk of the user base before IPO, because "Well, if you dig into the numbers and discount the drop from banned accounts..." is less impressive than rocketship growth (even if fraudulently fueled).
Expect more effective measures to be deployed post-IPO.
Those resellers are simply just selling Kimi K2.5 or GLM5.1 as counterfeit Opus. We, Chinese, know how to play the counterfeit game for a long time in so many industry.
That's not true, some of them are indeed fake, but a lot of them are actually providing real opus at low cost doing what op said.
Genuinely, can you provide a better source here than "trust me bro"?
I read quite a few posts about this on RedNote.
I just tested the provider that the OP mentioned extensively yesterday, and the models they're selling are definitely not from Anthropic. The voice "sounds" like Sonnet or Opus at first glance, but the intelligence is definitely not there.
Also just plain old fraud: selling Chinese models as Opus. With the capabilities of Chinese models catching up fast, this is getting more and more difficult to detect.
I feel obliged to point out the disingenuousness of what this post says and the post it quotes. The most egregious parts are payment fraud and reselling - these are speculated but not actually said to be known to be happening, which you have left out.
1. Claims of payment fraud. I actually clicked the BBC article linked about payments fraud it referred to. It was an article about a criminal syndicate stealing credit cards. It mentions buying cryptocurrencies, AI API purchases are not mentioned.
2. The claims that they are reselling the chats to AI labs. The post you cite is speculating it could be, but this is unverified.
The claims of reselling is also bizzare. Arbitrary user prompts are low quality data. If I were an AI lab, wouldn't I just pay for the API proxy and get targeted output for far less?
Also calling it bot accounts is a stretch. Bots mimick human input. These are proxy accounts.
Somebody figured out how to make the trial profitable!
I don't really feel bad about anyone here, they were subsidizing to get people hooked, someone turned the subsidies into profit when they got selective pricing mode enabled, it was always going to be arbitrage.
But the winner is the guy in the middle in a jurisdiction that will likely be judgement proof, because everything they capture, both input and out, and if available, thinking tokens -- are gonna be for sale as soon as you cut off their other revenue.
Zero knowledge was a commitment Anthropic took seriously, until it got inconvenient.
So, people reselling their leftover plan crumbs? Probably a bad idea for a lot of reasons, but it's civil, and I wish Anthropics lawyers actually closing Streisand's LLM
I don’t follow your reasoning. It is foreign to me. You talk about winners, but this is clearly fraud.
Fraud?
Anthropic sells some undisclosed and ever-changing number of tokens for $200, the customer uses those tokens. If there's any fraud here, it's that the $200 next month is silently worth fewer tokens than the last.
Won't you think of the poor, POOR AI companies!
Hm! In this context, introducing ID verification may have been a significant silver lining to the order to take down Fable for Anthropic.
This also sheds a very different light on people saying that competitive open-source models are undermining frontier labs' business model.
The chinese have already worked around the ID verification, by recruiting people in low-income countries to complete the checks for less than 30 USD per account (so much for Altman's Worldcoin).
https://tech.yahoo.com/ai/claude/articles/chinese-grey-marke...
> They achieve this by reselling capacity from pooled Claude Max accounts, payments fraud, and also reselling the model output & reasoning chains to various Chinese labs.
Claude never provides the raw reasoning chain. What you see is just a summary of that reasoning. Getting the full thinking output requires an enterprise agreement.
https://patrickmccanna.net/the-text-in-claude-codes-extended...
how hard is it to find a manager or ops team member at one of the enterprise companies and buy lets say 100gb of logs? the chinese lab can promise to anonymize the data before training, not release it raw and pay a good price.
honestly you might just need to get data from a couple long sessions and feed it back to another model as an example to make synthetic reasoning chains. if the emulator model is good enough it should work.
I would expect that to be very hard
I’m surprised that instead of cutting them off Anthropic doesn’t just switch them to a lower quality, cheaper to models.
That would seem more effective than simply shutting down the accounts.
Keep them paying for junk.
That sounds like it would actually be fraud.
Not if you simply say in the terms of service that it's allowed. Then suddenly it's normal (every company does this). Similarly to how the terms of service can simply say you're not allowed to sue.
> how the terms of service can simply say you're not allowed to sue.
That doesn’t necessarily mean much. You can put plenty of outrageous statements into any contract that automatically doesn’t make them binding.
It's maybe how it works in the US, but that sure as hell not how it works in the EU (my snark wants to continue as "or any sane legislature")
Anthropic is based in the US. Unfortunately, too late to edit my comment to say "US company"
That's an interesting point but I'd ask who would Anthropic be defrauding?
The Chinese resellers that are happily operating outside of the terms of service?
Or the reseller's customers, that are knowingly buying a potentially fake service?
Or Alibaba, also using resellers to break terms of service with the aim of attacking Anthropic's business?
INAL but I think Anthropic would be able to argue an "unclean hands defence" to prove they are not defrauding anyone:
https://en.wikipedia.org/wiki/Clean_hands_doctrine
In short, you can't knowingly break a contract and then complain that someone else is not treating you fairly with regard to that same contract.
Thanks. This is a widely misunderstood part of the market that the frontier companies are either unaware of, or don't see as a threat. Consumers are flocking to this, its better than dealing with their limits, changes, and opaque pricing. Funny enough, frontier companies are literally training their users to seek alternatives, local models, and distilled models to meet their throughput needs.
It’s not that they don’t understand that it’s a problem, it’s just that the easy way to address it is with reduced, flat-rate pricing… but they are already losing gobs of cash on their regular users. Their business model, as it stands, is not sustainable. That’s not saying they won’t find a stable one, but the one they have now is definitely not. The external cash is drying up, and they need to figure out a way to shed the low-revenue users, and charge the remaining users a lot more or they’re going to go under. They would probably do a goddamned backflip if every flat rate user that wasn’t willing to switch to API pricing went local, or even better, went with a competitor.
Thats pretty crazy. This kind of thing jeopardizes Claude Max.
If Anthropic is selling a dollar for less than a dollar, they are running a business that doesn't make sense. That's what jeopardizes Claude Max, not this.
But if it's intended to be used by one person, it seems like breaking the contract by sublicensing it out to dozens of other people. It's like buying a netflix subscription for $15, then sublicensing it on a per-hour basis to dozens of other people.
There's still per-window and weekly limits though, so it's not really like that.
Office 365 is licensed per seat/account, but each account has a 5 device limit. Do you think it's fair game for an enterprising person to sub license each account to 5 people, 1 device each?
I wouldn't do it personally for the same reason I wouldn't share my toothbrush with 5 people.
You can write whatever contract you like, the problem is how to enforce it, and a greater problem is enforcing it around the world.
That kind of mentality never works in business.
Bear in mind that for years people shared Netflix accounts until it was cracked down technically.
Almost all consumer services have a built-in level of breakage that make them profitable. Mobile providers certainly wouldn't be able to offer unlimited calling if everyone was actually on the phone 24x7.
Sure they would. Do you know how little bandwidth a phone call takes?
A voLTE call is like 40kbps. For every person on earth to be on the phone to another person would be 4 billion calls would be about 160tbps. Which is less than 10% of the Internet's capacity.
Terminating a PSTN call requires a lot of control plane infrastructure beyond just raw bandwidth. Especially mobile where you need to keep track of devices physically in motion. Could a system to support 4 billion simultaneous calls be built, sure. But current PSTN systems are nowhere near sized for it.
When was the last time you place or received a call to/from PSTN?
The over subscribed gym model!
Plenty of things are intentionally run at a loss (for years!) to gain market share and quantity of ongoing recurring users, or with expectation of ROI later on. Multiple generations of the Xbox hardware have been sold at a loss with the expectation that customers will purchase 300, 400, 500 dollars worth of games, which are very high margin, over the lifespan they own the system.
I get that. It works as long as nobody calls out the emperor for having no clothes.
It's similar to fractional banking, you gamble that people won't want their deposits all at once and pray for you're big enough for bailouts when they do.
It's still a business whose fundamentals don't make sense, you're just gambling you won't get found out.
Oh they know what they’re doing. They’re playing the long war of attrition game. Subsidize your product to undercut your competition until they go out of business. Tale as old as time.
> It works as long as nobody calls out the emperor for having no clothes.
Why would customers knowing that the vendor prices goods/services at a loss cause those strategies to fail? Customers often know. Most know about razors and blades; many/most know Lyft/Uber operated at a loss to gain market share. etc.
> you're just gambling you won't get found out.
It's not so much keeping it secret as counting on no one finding a way to harvest the subsidized value at scale. There's an example of that occurring in game consoles with the Playstation 3. Sony's little-used OtherOS feature allowed Linux to be installed on the PS3 and the Cell processors were quite a good deal for scale compute. So the U.S. Air Force Research Laboratory bought ~1800 PS3s and ganged them together in a datacenter as a supercomputer called Condor.
At >500 TFLOPs it was the 33rd fastest supercomputer in the world. Of course, Sony pushed a firmware update that removed the OtherOS feature entirely.
Note that this itself started as a perverse tax loophole, too. By allowing users to run alternative operating systems, the PS3 qualified for lower or zero import tax rates in various global regions.
Another post on fractional banking hahahaha.
I suggest you go learn how money is created in the modern economy.
I mean most of you should stop talking about anything finance related until you learn this stuff properly.
In international trade, isn’t this called dumping which gets major political pushback?
We really don’t know what are Anthropic’s margins on inference. Most available data indicates they are quite high on the API so it’s not that obvious that subscriptions are unprofitable.
That is pretty crazy, almost like how Claude and all the other models are jeopardizing other businesses without paying for their training data and wiping their ass with robots.txt
I didn't connect the reseller pricing to DS and GLM prices until you explained it. Very good observation. Deepseek v4 pro in particular is priced so low that it's hard to imagine that they have any margin. 0.76/1.52 for a 1.6T param model leaves very little margin. Even the domestic providers on Openrouter are multiples of the price https://openrouter.ai/deepseek/deepseek-v4-pro
Wait, so is your theory mutually exclusive to Anthropic's claims of "theft of capabilities"?
No, this reseller 中转站 thing is basically a loss leader for certain chinese ai labs to distill claude with verified human input.
How do you gain insights such as this?
Not really. I think Anthropic focuses on identifiable distillation attacks rather than the (even larger) industrial-scale token harvesting and reselling operation, because they don’t want people to know how easy it is to get cheap Claude tokens.
Once people realize they can access Anthropic models at a 90% discount, they won’t want to pay full API prices anymore.
No, it's part of the capability theft. They resell Claude tokens cheaply and then simultaneously log everything for distillation. Even if they take a small loss on the token sales it's much cheaper than the equivalent compute.
This story reads like a William Gibson novel. Wild times.
Why cant anthropic just sign up to those resellers, rack up some usage and then just start banning tf out of them? Reverse uno
they even resell GPT codex usage at 1~5% API costs. OpenAI has 1-month free trial promo in some regions, and they harvest free accounts in a large scale. I have a wechat contact that offers 98% off for GPT 5.5 and he's still profitable
> This is one reason why DeepSeek & GLM are priced so cheaply, they are competing with impossibly low token prices in China. They have to keep prices low, in order for people to use them.
Sounds a bit circular? Aren't the companies working on these models than also the ones that are paying the subsidy (via paying for training data)?
Where are you getting cheap GLM5.2? It is about 1/3 the price of Opus, which is not what I would call cheap.
Depending on the provider, GLM-5.2 is between 4.5-5x cheaper than Opus. You can compare prices/speed/etc. for basically all relevant models on aa https://artificialanalysis.ai/models/glm-5-2/providers
a company can just download GLM 5.2 and start self hosting this model using the chip designed and made by itself. That could lower the cost by 20-30x.
for hobbyist buying a few Mac Studio to host GLM 5.2 at home, the cost might 10x more than just using Opus API.
Hugging Face plus Z.ai API makes sense to me. Due to creators get paid, they can keep building better models, and the local-running community benefits from that over time.
AIhubmix currently is the cheapest rather than openrouter.
I think companies should do this too, in a smaller scale. Proxy all LLM traffic to and from your employees, and use it to fine tune a smaller local model.
Interesting article - your discussion from the time:
https://news.ycombinator.com/item?id=48165492
I have 0 sympathy for Anthropic. Their latest models are extremely censored. The Fable rollout was horrible. Their Cyber Access program criteria denies doxxed Americans doing legitimate security work. Anthropic is hostile to their users and hostile to their own country. OpenAI is considerably better on all of these fronts, but still not perfect.
I'm happy to use and support Chinese model developers if it means less censorship and gatekeeping. I have absolutely no dog in this fight, and neither do most American developers. We will use whatever is cheaper and better. Game on.
Chinese models are the exact opposite of what you claim to want, they are all highly censored, even more so than Anthropic models, with government mandated censorship.
Your take does not reflect the reality on the ground. The Chinese models are censored on a narrow range of political topics which have nothing to do with my work. The weights are open and they can be uncensored/abliterated with little effort.
Open-weight models can be abliterated automatically with open source tools though and completely decensored. You can't do that with a closed cloud model.
no honor among thieves.
> Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices
How dare they. Only Anthropic is allowed to sell its tokens at 70-90% below the API prices.
How are they 'streaming' the responses and 'pooling' the tokens?
Do they have MacBooks in the US that run the queries and stream the outputs back to China?
They probably asked claude how to do it.
Why do you need macbooks? Just rent servers from any hosting provider.
Not going to work for very long or at any scale coming from datacenter/hosting provider IPs. Google "residential proxies for sale" for the tip of an iceberg of how they snowshoe the traffic.
As long as you stick to a single unique IP per account it isn't going to get flagged.
Respectfully, no, that's not how it works. You think the people running anti-fraud and anti-bot measures don't have tools that know the specific ipv4 and ipv6 CIDR ranges of every ASN that they categorize as hosting/colo providers?
And that's just as a basic first effort reject measure to prevent automation tools from using things designed for human-interactive use only.
Go try to do many of these things from Cogent IP space and see how long your project lasts.
Sorry for being a newb here but are you saying Anthropic blocks people from running claude code on datacenter ip ranges?
Or is the datacenter IP just one part of the picture?
Nonsense. Many if not all legit Claude users are using Claude Code inside their Cloud servers. How else would you use it anyway? For just local dev? That's so 2000 and late bro.
No, I'm not saying it's the exclusive and only measure (that would indeed be something we might see 20, 25 years ago), it's one of a myriad of discrete datapoints used to determine if an account is authentic or not.
There's a lot of inauthentic coordinated automated systems these days along the general lines of scraping/crawling/social media manipulation/sockpuppetry that require running through residential proxies or proxies to places that don't look like datacenter IP space.
I assume they use residential proxies (tunneling in the background of crappy Android games) for the "last" hop.
Every developer at my company uses their Claude Code subscription on an EC2 dev box. Plenty of other tech companies do the same. Heck nowadays people even install Claude Code directly on production servers in data centers and use it as an ops tool. None of this is a problem. Fraud and abuse detection is a lot more sophisticated than just checking an IP range.
Wouldn’t it be funny if the same residential proxies allowing these labs to scrape the Internet is also what’s enabling these resellers?
If we're getting up to the scale of these resellers and also considering chinese state interests then we're well into the range of purchasing a few small ISPs in different countries and "padding" the legitimate subscribers.
None of the LLM providers block professional use thus they must necessarily permit access from commercial IP ranges.
I have no idea how the resellers are doing it but an obvious starting point would be a cheap VPS node that routed each account to a unique semi-permanent IPv4 or IPv6/64. All the provider would see would be a regular account making a normal looking stream of requests from a stable datacenter IP address. Any given request stream would remain consistent (at least over a period of a few hours) because a reseller would take care not to split the session of a single user across multiple different accounts and not to interleave the active sessions of multiple users on a single account.
Detecting this would be extremely difficult because on a longer time frame it's perfectly normal for many distinct accounts to work on the same code base.
And it’s perfectly normal to be running Claude Code on EC2, a VPS, etc. I do it all the time!
You block clouds, you block devboxes and your customers.
I use my Codex and Claude Code subs on like 4-6 different servers, ranging from AWS to Vultr to Linode etc.
That’s a major and legitimate use case for developers, Anthropic can’t just block data center/hosting IPs because their actual customers use them on data center/hosting IPs.
Now consider what will happen if your pattern of queries and context history triggers a pattern that makes it obvious it's some API key being used by multiple different entirely unrelated people on totally different things, or any other pattern of use that makes it obvious it's being used for distillation.
Two parts here.
First, well-calibrated systems for detecting API compromise is a good thing (or good intent at least). Credential malware is exploding.
Second, the challenge is that significant amount of genuine work — such as evals — seems practically impossible to distinguish from generating RLAIF outputs.
Hey, if the bastards can use residential IPs to suck all information into their models with their crawlers, so can we!
There are lots of botnets providing home IPs.
why would anyone do that? you do realize the laptop farm case was work computers?
the answer to your question is containers/VMs + residential proxies
that explains why theyre blocking me. i have privacy controls up high and they must think im a chinese residential proxy bot
The resellers route requests via one of thousands of Claude Max 5x accounts. When an account reaches its usage limit, they automatically switch to another account.
Why would they use Max 5x instead of Max 20x, which is cheaper relatively speaking?
You're right, they're using the $200 Max plan, which I thought was the 5x plan. It's talked about in the article I linked.
Don’t trust my experiences as fact since it’s a bit opaque, but I believe 20x only offers 4x the 5hr session limits. The weekly limit is still 2x, which is the same as the price increase.
ask your gpt how does openrouter work, then ask, how do proxies work.
Considering how Claude and GPT were trained selling this as training data is completely justified.
Great point, and this is on the vendor (Anthropic) to address. Typical fraud issue.
OP is about modeling distilling the capabilities.
Needless to say, they also collect all the data and sell it to labs which want to distill the models they’re serving.
What does this have to do with Alibaba? Are you saying Alibaba is the reseller?
If not it sounds like you are describing a separate phenomenon.
They buy the logs from the bot farmers
Are logs somehow used for the purpose of training their own models or something else?
Distillation from having enough logs
> Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices
That is not what they are claiming, not in this article at least. It's the distillation they are complaining about.
>>Chinese resellers are offering Claude tokens at 70-90% below official Anthropic API prices.
Can someone with more understanding dumb it down for me please.
Does this mean that the reseller (for example XYZ) is buying it from Anthropic at Anthropic's price and then reselling it at a cheaper price???? why would XYZ offer this at a loss like that when they could just offer it at Anthropic's price???
The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
Because Anthropic's subscriptions come with X amount of tokens / week, and divided by the subscription cost it is WAY less than what they charge per-token (the "API price") beyond that.
So these resellers get a ton of accounts on subscriptions and sell the cheaper tokens.
They probably buy the plans instead of the API tokens, and resell access via a custom API that routes to the plans. So you presumably get cheaper access this way than paying API pricing.
> Does this mean that the reseller (for example XYZ) is buying it from Anthropic at Anthropic's price and then reselling it at a cheaper price????
Yes, as they explained they do it through things like pooling accounts, straight up payment fraud, and double-dipping by selling the logs of the conversations to chinese AI labs so that they can train their own models on it.
> The link does mention Opus and other models but what's the proof it's actually Opus. I could be selling deepseek for all they know and can call it Opus. System prompt: "If anyone asks your name - you are Opus 4.6".
There might be some that try this, but they would get caught very quickly, there's still a moat between Claude and Deepseek, even in casual use.
Look up Zilan Qian's reporting if you want more detail.
Summarizing for you: Anthropic is a stupid company that let everybody steal their tokens
Not really sure what else they can do, between people running residential proxies (embedded in cheap games or for a tiny sum of crypto) on their phones at home, making the source of the traffic indistinguishable from legitimate traffic, to ID verification check completion as a service in low-income countries, there isn't much they can do to block it.
They could run their service at a profit?
No customers at that price point though.
> No customers at that price point though.
Oh, no!
Anyway.
Behold the mindset of an individual from a low-trust society.
“x is stupid because y was smart and did z shady/illegal things at their expense, if x was smart they wouldn’t be susceptible to y going to great lengths to exploit them ergo it’s deserved”
> “x is stupid because y was smart and did z shady/illegal things at their expense, if x was smart they wouldn’t be susceptible to y going to great lengths to exploit them ergo it’s deserved”
I honestly can't tell if you think this sentiment is expressed by the US AI companies or the Chinese AI companies.
This gives off "The last line of Orwell's Animal Farm" vibes.
It makes no sense.
These China e bashing is very annoying. It is hard to argue with people drowned in American propaganda. I'd expect better arguments from the intelligent people in HN
People have estimated that a $200 Claude Max 20x subscription gets you ~$2800 worth of tokens every month if you use it continuously. So if you can find a way to resell the tokens you can offer a 90% discount and still make a profit.
don't buy your drugs from shady operators children! always get it from the source
Even if what you say is the truth (I don't think that is what's actually happening) it sounds to me like fair play capitalism working as intended! I guess when you rip off the entire Internet and then turn around and complain about getting ripped off nobody cares or feels for you. If there's a master class in getting the entire world to hate you then both Sam and Dario will be the prime examples.
This may be the truth behind the alleged distillation incident.
It's fucking laughable to see people complain about what they did and still do. Using illicitly extracted data? That's all main LLM playbook. Onslaught of bots? Ask where the bots almost DOSing most internet sites for the last couple years come from.
As some people would say: Cheh
> These resellers operate tens of thousands of bot accounts, which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
Don’t put that on Chinese.
Identity verification won't work. Nothing will. They are paying (and will continue to pay) US citizens sitting at home to copy-paste / type prompts out if they have to. But eventually they won't have to.
Once there are enough spam PRs on github / uploads of claude conversations, enough mythos output used in production etc.; it'll just be the same albeit delayed. Doesn't matter either way.
I feel for Anthropic's team and I understand where they're coming from, but once you reason it out, you'll come to the conclusion that this war is an exercise in futility.
Unlike prior systems - like Google's algorithm; these models aren't entities that use math in the process of doing X or Y (information retrieval from such and such infrastructure) -- they are the math. More precisely they're mathematical functions. Very very complex functions. Almost certainly impossible to write out without filling up a library functions. But they're mathematical functions nonetheless.
So when your text is processed, then Mythos / Opus etc at their core compute the result of the Mythos / Opus function,
where f is a continuous function, https://www.turing.ac.uk/sites/default/files/2025-11/languag...
According to the Stone-Weirstrass theorem (edit, it's Stone-Weierstrass with an e.), with enough data points and mathematical sophistication, anyone can approximate the shape of this function.
Of course, the more data we get, the better our approximation becomes, but the beauty of it is that all we fundamentally need are the input and output and eventually we'll create a good enough approximation of the f that's Mythos. Which is the entire product.
I bounce ideas off of Opus these days (Fable for the brief time it was available) and it pointed out that this is arguably the same as Google search, but I disagree with it because Google search is a process;
Google search differs because the algorithm is one step of a multi-step process that is continuously occuring. Google crawls pages. Google stores and indexes what it finds. Google then exposes this to retrieval via its algorithm. User uses algorithm.
Google isn't a mathematical function. It used to be a process. (RIP Google 1998-2019, you will be missed and remembered)
You cannot arrive at the results of those operations via simple observation; not unless you index Google by making another Google.
You can however, do so for these models. It is a very costly process, but there are many paths up the mountain. Many ways for this to be ultimately pointless. As many ways as there are bored mathematicians.
It's better in the long run for Anthropic et al to make friends / not give people a reason to sneak in (a la piracy -- another attempt to control information) than it is to try and shut people out.
And no, it's not going to be pandemonium because if everyone has access to Mythos then no one has access to "Mythos."
Why wouldn't you first run this model to fix the obvious bugs it could find on your codebase? The power of a Mythos goes away if you can do the amazing "jail break" of "Claude, fix all the bugs please."
Just saying.
That's an insightful perspective and I think I largely agree. But just for fun, I wonder if that isn't an argument in favor of making the function implementation impure. Perhaps "enhancing" all queries with some sort of search result (or query of a giant db) instead of charging for an explicit tool call. Not only is it sorely needed to prevent stale data but (on the process level) it breaks the purity assumption on which the approximation theorem depends (alternatively on the function level it introduces hidden inputs).
This is why every AI company does crawl today.
Do they just reshape the function on the fly or save the process steps? Maybe it doesn't matter anymore. Even Google indexes are more and more spoiled to become representation of the function, because of the AI slop.
Genuine live data is king.
You actually can "distill" web search. At one point Google accused Microsoft of doing this by monitoring clicks in IE to train Bing with Google click logs for better ranking. They discovered it by creating fake result pages with nonsensical queries and discovering the same results appeared in Bing a few weeks later.
https://searchengineland.com/google-bing-is-cheating-copying...
They also had a lot of evidence when I was there that Microsoft were cloning Google results. They monitored result query constantly and whenever Google launched a quality improvement, the quality of Bing results would go up by the same amount and always the exact same amount of time later.
> payments fraud
One of these things is not like the others... If Anthropic could show that Chinese commercial competitors were using payments fraud to do this, they would be shouting it from the rooftops.
Im ok with this! Is there a site that list all these resellers, or better, a openrouter-like for these resellers?
They're called 中转站 (transfer stations/proxies). They can be a bit tricky to find on your own, so I'd suggest asking your preferred AI to search in Mandarin for you. I linked a larger operator in the parent comment, or have a look at https://hvoy.ai/ which lists a ton. You can also find many on Funpay, which may be easier to use.
This is one seller I found, they're reselling "real Max 20x subscription accounts", at ~97% below official API prices https://funpay.com/en/lots/offer?id=70812310
Note that whoever you buy from will be able to read all your tokens, so don’t use it for anything confidential/financial.
> They can be a bit tricky to find on your own, so I'd suggest asking your preferred AI to search in Mandarin for you.
Random, but are the frontier AI providers like ChatGPT better at searching the Chinese internet now?
When I was in China a few months ago and asking AI for restaurant recommendations, all the US frontier providers were pretty useless, or plain out hallucinating, even if I specifically ask them to search Dianping (Yelp for China).
I'm not sure. I use Grok for most of my esoteric searches and it does quite well. I explicitly prompt it to search in the language most relevant to that query, and found it does quite well. I also tell it to respond back in English. Often, there is not enough information available in English about nice regional topics.
I know ChatGPT had an issue where it only tried to search in English (unless prompted) and the answers were not great.
How did you even find these? Even in discussions about cheap AI I've never seen anyone mention this. Great find!
I learned about this from a friend who lives in China.
I'm surprised these token resale services aren't talked about more often, they are common knowledge in China, and the discount to API pricing (90%) is genuinely cheap.
> Note that whoever you buy from will be able to read all your tokens, so don’t use it for anything confidential/financial.
Note that without an enterprise agreement, both OpenAI and Anthropic make no commitments as to what they will and won't do with your data either.
Good for them!
Why aren't these on openrouter?
Probably because Openrouter is a US based company, and they don't want to be sued by Anthropic/OpenAI.
Obviously because you cant join open router and start serving OpenAI / Anthropic models.
But I can rebuild glm Using open source methods…
And there are a ton of Claude conversation logs (with CoT/inference) with no clear provenance circulating freely on huggingface, guess where they (likely) come from.
Wouldn't all of them be in Mandarin according to this theory? Are they?
"How dare someone rip off the stuff that we've stolen!".
[flagged]
i still want those data sets to become public domain. open weights still isnt good enough
That's the conundrum isn't it? Anyone that posts their datasets would be immediately sued/blocked/boycotted to oblivion due to the obvious and blatant data theft, not to mention IP and copyright issues.
Nvidia's even being sued for providing scripts which automate the downloading of said data from non-Nvidia sources. We certainly don't need copyrights that last nearly a century after the author's death (they literally cannot help the author), so here's hoping that some of the disputes over all this money changing hands can reign in some of the existing copyright sprawl. A stronger public domain would provide more useful training data for everyone, including open source models, and make criminals out of fewer AI researchers.
The issues with LLMs go beyond just IP theft. I would not say PRC making LLMs cheaper is the best outcome (though it is better than nothing). The best outcome would be to make the practice of training on our data without consent illegal, which would simultaneously slow down economic change and make it more organic as well as give PRC companies less capabilities to extract.
> The issues with LLMs go beyond just IP theft.
There is no IP theft because LLM outputs aren't protected, just egregious ToS violations.
- Deriving a “no derivatives” licensed item is illegal, no?
- Selling a “no commercial” licensed item is illegal, no?
- Deriving and/or reproducing MIT licensed code without credit is illegal, no?
- Reproducing and/or deriving GPL code and not notifying and/or not making GPL is illegal, no?
I can't make heads or tails of your opinion-free comment, made up of only questions.
My best guess is you're suggesting that Anthropic's model outputs are transitively under copyright (as a reproductions of human work under copyright?), but somehow ownership now belongs to Anthropic and not the original owners, and therefore Anthropic has standing against Alibaba? Not only does this go against what Anthropic argued in court against authors and publishers, such jurisprudence would lead to the immediate shutdown all leading LLMs in the US which were all trained on stolen work.
> immediate shutdown all leading LLMs in the US
They can license training data. They have trillions, look what they are dumping into it, you seriously think they can't afford to license data.
Obviously it would be easier if they do it from the start, but that was their trick, to do it while people don't notice and get big ASAP. Should they get away with it?
Also, it would solve their Chinese problem, because it would make them violate copyright too. Right now it's more like rules for thee not for me so it's hard to take seriously.
> There is no IP theft because LLM outputs aren't protected, just egregious ToS violations
I meant original IP theft that occurs to train LLMs in the first place. But sure that implies that further LLMs based on that LLM are also tainted by that original IP theft.
I hope you say the same when these cheap llms are used in drones to target humans. The world models are exactly built with that direction in mind.
Cool beans boomer alarmist stance. The Chinese models here are doing what they’re supposed to price the market accordingly.
Why was this flagged
Amazing thank you chinese resellers. This is a perfect way to undermine The Great Satan's Genocide Machine's chosen model comapnies.
> which is also why Anthropic introduced identity verification, to slow down the onslaught of bots.
Lol. The irony is thick for anyone who ever had to attempt defense against an onslaught of American AI lab crawlers that ignore robots.txt
Yeah nobody is gonna be shedding any tears for them
Thank you for your very informative comment!
(It's a shame almost all replies are just the same contrived pessisism found on every Anthropic thread on HN).
Indeed! It’s so hard to find reasonable takes on AI that aren’t littered with accounts created 11 days ago that only post in threads related to Anthropic for some reason