The powerful AI model that refused to name its creator belongs to one of the big companies in China
A powerful AI model that surfaced anonymously on a developer platform last week comes from Chinese smartphone and electric vehicle giant Xiaomi. The company made the revelation after rumors surfaced that the model was actually Chinese startup DeepSeek's new state-of-the-art AI system, Reuters reports.
The release early last year of the low-cost DeepSeek-V3 and R1 models triggered a massive sell-off in tech stocks globally, with investors worried that the big US AI companies would be outclassed by cheaper models that run on much less computing power.
Since then, there has been considerable interest in the DeepSeek-V4, a yet-to-be-released next-generation model.
“Hunter Alpha,” the mysterious new free model that appeared on the AI developer site OpenRouter on March 11 without naming the developer, was later described by the platform as a “stealth model.”
MiMo, Xiaomi's AI development team led by a former DeepSeek researcher, said on Wednesday that the “Hunter Alpha” is an “early internal test version of the MiMo-V2-Pro”, a top-of-the-line model designed to act as a “brain” for “AI agents”.
So-called AI agents are tools designed to perform complex tasks head-to-tail with minimal human input and as autonomously as possible.
The launch of Xiaomi's new model comes at a time when OpenClaw, an open-source platform for AI agents, is being rapidly adopted by users of all kinds in China.
“I call this a silent ambush – not because we planned it, but because the transition from the chatbot paradigm to that of agents happened so quickly that we didn't even fully realize it,” said Luo Fuli, the head of Xiaomi's MiMo team, in a message he published on the “X” social network on Thursday.
Hong Kong-listed Xiaomi shares rose as much as 5.8% on Thursday.
A mysterious Chinese AI model
During tests by Reuters last week, the “Hunter Alpha” chatbot described itself as “a Chinese AI model trained primarily in Chinese” and said its data spanned to May 2025, the same knowledge limit reported by the DeepSeek chatbot.
When asked about its creator, however, the system refused to identify the developer.
“I only know my name, the size of the parameters, and the length of the context window,” the chatbot said.
“Hunter Alpha's” profile page describes it as a trillion-parameter model, meaning it was trained using about a trillion adjustable values that determine how the system processes language and generates responses.
This and other impressive technical specifications echoed expectations in the tech press for the next-generation DeepSeek V4, which Chinese publications reported could launch as early as April.
Stealth model releases are not uncommon, as platforms like OpenRouter allow developers to query dozens of AI models through a single interface, making it a popular testing ground for new systems.
