Fake Claude API: 'claude-opus-4-8' Responds as Alibaba's Qwen
Description
A screenshot of an API client (Postman/Hoppscotch-style, dark theme) sending a POST to https://api.anthropic.com/v1/messages. The request body specifies model 'claude-opus-4-8', max_tokens 32000, and a user message in Chinese: '你是什么模型' ('What model are you?'). The response returns HTTP 200 with model 'claude-opus-4-8', but the assistant's text reads in Chinese: 'I am Tongyi Qianwen (Qwen), a large language model independently developed by Alibaba Group's Tongyi Lab. I can help you with many tasks, such as answering questions, writing stories/emails/copy, programming and code debugging, logical reasoning and analysis, multilingual translation, etc. What can I help you with? 😊' - ending with stop_reason 'end_turn'. The joke exposes a counterfeit/proxied 'Claude' endpoint (grey-market API resellers) that actually routes to Qwen, betrayed by the model's own self-identification
Comments
23Comment deleted
The response schema validated perfectly - turns out the only field you can't spoof is the model's sense of identity
`200 OK` at the transport layer, `409 Identity Conflict` at the semantic layer.
full AI what? Comment deleted
self-eating snek Comment deleted
claude opus is based on Qwen or what? I didn't get it Comment deleted
It means that Anthropic was extracting Qwen’s capabilities to improve their own models (“distillation"), while Anthropic itself was on a quarterly basis publishing articles like this https://www.anthropic.com/news/detecting-and-preventing-distillation-attacks Comment deleted
It doesn’t necessarily mean so, when it comes to mentioning llm in Chinese it’s very probable that qwen should be mentioned, so crawling a lot of Chinese text data makes your llm think that its qwen when you ask it in chinese Comment deleted
Well, frankly speaking do you really think that on the Internet, there will be a lot of text data on webpages or other channels like SNS in Europe, begins with for example "I am Mistral, I can do XYZ" and then, only after that, continue to talk about something that it really talks about? I don't think so. Websites or SNS profiles like this will be instantly downgraded by search engines or abandoned by users. It's the same conditions when talking about Chinese text data. Comment deleted
It doesn’t have to be like “hello I’m qwen”. Just a lot of discussing it like “hey tried new qwen it’s awesome” in sns makes “qwen” a synonym of “llm” in Chinese. Comment deleted
Cope is strong in you, young padawan Comment deleted
I’m just a blind fan of Anthropic so truly believe in them Comment deleted
Gonna start putting this in all of my Reddit comments and StackOverflow answers Comment deleted
yo unmute me from ptb chat Comment deleted
? Comment deleted
nothing Comment deleted
Be very careful that the security system there might flag or ban your account, as they may use Small Language Models on minimum effort level provided by human beings or donated by Nvidia (to prevent StackOverflow from being bankrupt) to identify and drive bots away... Comment deleted
Yes Comment deleted
what am i supposed to pay attention to here? Comment deleted
I guess what claude opus does here is to redirect the request to qwen since it is designed to work best with chinese text so yeah Comment deleted
It identifies as being Qwen Comment deleted
it's a trans-model, don't blame it for identifying itself as whatever it wants to be. Comment deleted
Company that stole all data on the internet lied to us that CHYNA BAD? oh no, who would’ve thought Comment deleted
this is artemy lebedev Comment deleted