Xiaomi Claims Mysterious 'Hunter Alpha' Model, Promises to Open Source MiMo-V2 Series

March 19, 2026 — The tech community's speculation is over. Xiaomi has officially claimed responsibility for the mysterious AI model known as "Hunter Alpha," which surged to the top of OpenRouter's API leaderboard earlier this month, sparking widespread rumors it was an early version of DeepSeek V4.

In an early morning announcement, Xiaomi introduced three new large language models under the MiMo-V2 banner: MiMo-V2-Pro, MiMo-V2-Omni, and MiMo-V2-TTS. The models are available for limited-time free trials for one week.

A 'Silent Ambush' for the Agent Era

Luo Fuli, head of Xiaomi's MiMo large model team, described the release in a detailed post as the company's "first true full-stack product series built for the age of intelligent agents."

"I call this a silent ambush," Luo wrote. "Not because we planned it that way, but because the shift from chat mode to agent mode happened so fast, it caught even us by surprise."

Technical Foundation and Rapid Pivot

The 1-trillion parameter base model began training several months ago with an initial focus on improving long-context reasoning efficiency. According to Luo, the hybrid attention mechanism developed represents "genuine innovation without over-engineering" and serves as an ideal foundation for agent systems.

Key architectural features include a 1-million token context window and MTP (Mixture of Token Pruning) inference for ultra-low latency and cost.

The Agent Revelation

Luo revealed that everything changed when she first experienced what she calls "orchestrated context" — complex agent frameworks. The impact was immediate and profound.

"I was shocked on day one," she admitted. "I tried to convince the team to use it, but failed. So I issued a strong directive: Any MiMo team member with fewer than 100 conversations tomorrow can resign or leave."

The tactic worked. Once the team's imagination was ignited by the capabilities of agent systems, their research velocity increased dramatically.

The Development Philosophy

Reflecting on the rapid progress, Luo shared insights from her experience building DeepSeek R1:

• Backbone and infrastructure research requires long cycles and year-long strategic determination.
• Post-training agility is a different capability — product intuition drives evaluation, iteration cycles shorten, and paradigm shifts happen earlier.
• Constant factors remain: curiosity, sharp technical intuition, decisive execution, and full commitment.
• One easily underestimated element: genuine love for the world you're creating.

Open Source Commitment

In a significant move for the open-source AI community, Luo promised that the new MiMo-V2 series will be open sourced. "When the models are stable enough to be worth open sourcing," she clarified, indicating a commitment to releasing robust, production-ready models rather than experimental versions.

The announcement positions Xiaomi as a serious contender in the increasingly competitive agent-focused AI landscape, challenging assumptions about which companies are driving the next wave of AI innovation.