Mystery AI model revealed to be Xiaomi’s following suspicions it was DeepSeek V4 

By Eduardo Baptista

BEIJING, March 18 (Reuters) – A powerful artificial intelligence model that appeared anonymously on a developer platform last week was revealed on Wednesday to be from Chinese smartphone and EV giant Xiaomi, after it fueled speculation that startup DeepSeek was quietly testing its next-generation system ahead of a launch.

The release of DeepSeek’s low-cost models DeepSeek-V3 and R1 triggered a global tech stock selloff last year, causing investors to question whether U.S. AI firms needed to spend billions of dollars on AI computing power. Since then, there has been a great deal of interest in DeepSeek-V4, a next-generation model that has yet to be released.

The mysterious free model, called Hunter Alpha, surfaced on the AI gateway platform OpenRouter on March 11 without any developer attribution and was later described by the platform as a “stealth model.”

Xiaomi’s AI model team MiMo, run by former DeepSeek researcher Luo Fuli, on Wednesday said that Hunter Alpha was an “early internal test build of MiMo-V2-Pro,” a flagship model designed to serve as the “brain” of AI agents, tools that can allow users to execute complex tasks with fewer human prompts and supervision when compared with a chatbot.

Xiaomi’s new model release comes at a time when OpenClaw, an open-source agent framework, is being rapidly adopted by users of all stripes in China.

“I call this a quiet ambush – not because we planned it, but because the shift from chat to agent paradigm happened so fast, even we barely believed it,” Luo said in an X post on Thursday.

“People ask why we move so fast,” she added. “I saw it firsthand building DeepSeek R1.”

MiMo-V2-Pro will partner with five major agent frameworks, including OpenClaw, to offer a week of free access to developers worldwide.

Xiaomi’s Hong Kong-listed shares surged as much as 5.8% on Thursday.

THE MYSTERIOUS CHINESE MODEL

During tests conducted by Reuters last week, the Hunter Alpha chatbot described itself as “a Chinese AI model primarily trained in Chinese” and said its data extended back to May 2025, the same knowledge cutoff point that was reported by DeepSeek’s own chatbot.

When asked about its creator, however, the system declined to identify its developer.

“I only know my name, my parameter scale and my context window length,” the chatbot said.

Hunter Alpha’s profile page describes it as a 1-trillion-parameter model, meaning it was trained using roughly one trillion adjustable values that determine how the system processes language and generates responses.

The system also advertises a context window of up to one million tokens, a measure of how much text an AI model can process or remember during a single interaction. A token roughly corresponds to a short piece of text, such as part of a word.

“The combination that stood out was Hunter Alpha’s 1-million-token context paired with reasoning capability and free access,” said Nabil Haouam, an engineer who builds AI agent systems.

“Most frontier models with that context window come with real cost at scale,” he added.

Those specifications resembled expectations in local media for DeepSeek’s next-generation V4 model, which Chinese outlets have reported could launch as early as April.

Umur Ozkul, who runs independent AI benchmark tests, said speculation connecting the model to DeepSeek was understandable given the timing and capabilities advertised.

STEALTH TESTING

Stealth model launches are not unusual, as platforms like OpenRouter allow developers to send queries to dozens of AI models through a single interface, making them a popular testing ground for new systems.

An anonymous model called Pony Alpha appeared on OpenRouter in February before Chinese firm Zhipu AI confirmed it was part of its GLM-5 system five days later.

A notice on Hunter Alpha’s profile page said all prompts and completions for the model “are logged by the provider and may be used to improve the model,” underscoring the industry-wide practice of using stealth model launches for unbiased feedback.

The model was adopted rapidly after appearing on the platform, surpassing one trillion tokens in total usage and topping the leaderboard charts on OpenRouter, according to MiMo.

(Reporting by Eduardo Baptista; Editing by Thomas Derpinghaus)