Xiaomi revealed as source of mystery AI model first suspected to be DeepSeek V4
The Chronify
A powerful artificial intelligence model that appeared anonymously on a developer platform last week and sparked speculation it was DeepSeek’s unreleased V4 system has now been identified as a Xiaomi model, according to Reuters and Xiaomi’s own AI team.
The model, called Hunter Alpha, surfaced on OpenRouter on March 11 without any developer attribution and was later labeled a stealth model. Its sudden appearance, strong reasoning ability, free access, and large context window quickly drew attention from developers, many of whom suspected DeepSeek was quietly testing its next generation flagship before launch. Reuters reported that interest was especially high because DeepSeek-V4 has not yet been released and remains one of the most closely watched upcoming Chinese AI models.
That speculation ended on Wednesday, when Xiaomi’s MiMo team said Hunter Alpha was an early internal test build of MiMo-V2-Pro. Xiaomi described MiMo-V2-Pro as its flagship foundation model built for agentic workloads, designed to act as the “brain” of AI systems that can handle complex, multi step tasks with less human supervision than standard chatbots.
Reuters said the confusion was understandable because Hunter Alpha’s profile advertised traits that matched market expectations for DeepSeek V4, including a one million token context window and a one trillion parameter scale. In Reuters testing, the model also said its knowledge cutoff extended to May 2025, the same cutoff associated with DeepSeek’s own chatbot, adding to the buzz around its identity.
Xiaomi’s official release provided fuller technical details. The company said MiMo-V2-Pro has more than one trillion total parameters, 42 billion active parameters, and support for up to one million tokens of context. Xiaomi also said the model is optimized for agent scenarios, with improved tool use, multi step reasoning, and long context performance.
The launch is notable for another reason. Reuters reported that Xiaomi’s MiMo team is led by Luo Fuli, a former DeepSeek researcher, and that the release comes as China’s AI industry shifts more aggressively from chatbots toward agent based systems. Xiaomi’s release also ties MiMo-V2-Pro to OpenClaw, an open source agent framework that Reuters said is gaining rapid adoption in China.
Xiaomi said Hunter Alpha’s week long stealth run on OpenRouter generated massive early interest, topping the platform’s daily charts and surpassing one trillion tokens in total usage. The company framed the anonymous test as an early validation phase before the full MiMo-V2-Pro rollout.
The episode shows how closely developers are tracking the next wave of Chinese AI models after DeepSeek’s earlier systems disrupted the market. Reuters noted that DeepSeek’s low cost V3 and R1 models triggered a global tech stock selloff last year by raising fresh doubts over whether leading US AI firms needed to keep spending so heavily on computing infrastructure.
That speculation ended on Wednesday, when Xiaomi’s MiMo team said Hunter Alpha was an early internal test build of MiMo-V2-Pro. Xiaomi described MiMo-V2-Pro as its flagship foundation model built for agentic workloads, designed to act as the “brain” of AI systems that can handle complex, multi step tasks with less human supervision than standard chatbots.
Reuters said the confusion was understandable because Hunter Alpha’s profile advertised traits that matched market expectations for DeepSeek V4, including a one million token context window and a one trillion parameter scale. In Reuters testing, the model also said its knowledge cutoff extended to May 2025, the same cutoff associated with DeepSeek’s own chatbot, adding to the buzz around its identity.
Xiaomi’s official release provided fuller technical details. The company said MiMo-V2-Pro has more than one trillion total parameters, 42 billion active parameters, and support for up to one million tokens of context. Xiaomi also said the model is optimized for agent scenarios, with improved tool use, multi step reasoning, and long context performance.
The launch is notable for another reason. Reuters reported that Xiaomi’s MiMo team is led by Luo Fuli, a former DeepSeek researcher, and that the release comes as China’s AI industry shifts more aggressively from chatbots toward agent based systems. Xiaomi’s release also ties MiMo-V2-Pro to OpenClaw, an open source agent framework that Reuters said is gaining rapid adoption in China.
Xiaomi said Hunter Alpha’s week long stealth run on OpenRouter generated massive early interest, topping the platform’s daily charts and surpassing one trillion tokens in total usage. The company framed the anonymous test as an early validation phase before the full MiMo-V2-Pro rollout.
The episode shows how closely developers are tracking the next wave of Chinese AI models after DeepSeek’s earlier systems disrupted the market. Reuters noted that DeepSeek’s low cost V3 and R1 models triggered a global tech stock selloff last year by raising fresh doubts over whether leading US AI firms needed to keep spending so heavily on computing infrastructure.
Related News
📚 Categories
You may like
