Xiaomi reveals MiMo-V2-Pro: Hunter Alpha was not a new DeepSeek model

Xiaomi has officially revealed the origins of the mysterious Hunter Alpha model, which appeared on OpenRouter on March 11, 2026, without any identified developer. It turned out that under this name, MiMo-V2-Pro was being tested — the company's flagship model for agentic scenarios — and it was precisely its characteristics that led the market to speculate it might be a new DeepSeek.

How the model was revealed

When Hunter Alpha was first spotted on OpenRouter, it immediately sparked a wave of discussion. The model identified itself as Chinese, said it had been trained primarily on Chinese-language data, and listed a knowledge cutoff of May 2025. When asked about its creator, it refused to answer, limiting itself to saying it only knew its name, parameter scale, and context window size. Against the backdrop of expectations surrounding DeepSeek V4, that was enough for rumors to spread across the market within days.

On March 18, 2026, the Xiaomi MiMo team announced that Hunter Alpha had been an early internal test build of MiMo-V2-Pro. The intrigue was compounded by the team's composition: the MiMo division is led by Luo Fuli, a former DeepSeek engineer. In her statement, she described what was happening as a sharp shift from ordinary chatbots to agentic systems — models that don't just respond, but can execute chains of actions through tools and external services.

"The shift from chats to agents happened so fast that even we could

barely believe it."

What MiMo-V2-Pro can do

Xiaomi positions MiMo-V2-Pro as a foundation model for real-world agentic workloads. The idea is that such a system should be not just a conversational partner, but the "brain" of an agentic infrastructure: coordinating complex workflows, maintaining long context, calling tools, and reliably completing tasks to their conclusion. The emphasis is therefore placed not only on overall language capability, but also on practical behavior in scenarios where the model needs to plan, choose steps, and act sequentially.

• Over 1 trillion total parameters
• 42 billion active parameters
• 1 million token context window
• Support for five agentic frameworks, including OpenClaw
• Free developer access for one week following the announcement

In the Artificial Analysis rankings, the model placed eighth globally and second among Chinese language models. Xiaomi also claims that in code generation, MiMo-V2-Pro surpasses Claude 4.6 Sonnet, and in overall agentic task performance approaches Opus 4.6. The company also highlights that the new version has seen notably improved stability and tool-calling accuracy compared to the previous lineup.

Why everyone thought of DeepSeek

The idea that Hunter Alpha might be a new DeepSeek did not arise out of nowhere. After the release of DeepSeek-V3 and DeepSeek-R1, the market became far more attentive to Chinese AI labs: their affordable yet powerful models significantly shook investor expectations and raised the question of whether the industry truly needed its previous scale of compute spending. Against this backdrop, an anonymous model featuring reasoning, a million-token context, and strong agentic performance looked like a perfectly plausible stealth test of DeepSeek V4.

The launch format itself also played a role. Platforms like OpenRouter have long served as convenient venues for quiet evaluations: developers receive a live stream of requests, while users test the model without pre-formed brand expectations. According to MiMo, Hunter Alpha quickly accumulated over 1 trillion tokens of usage and rose to the top of OpenRouter's rankings.

After Xiaomi officially revealed the model's origins, its Hong Kong shares momentarily gained as much as 5.8%.

What this means

The story of MiMo-V2-Pro shows that the new competition is no longer centered only around chatbots, but around agentic models capable of using tools and executing long chains of tasks. For Xiaomi, this is a bid to be seen not merely as a smartphone and electric vehicle manufacturer, but as a provider of its own AI platform. For the market, it is yet another signal: powerful models can now come not only from startups like DeepSeek, but also from large consumer-facing ecosystems.