Xiaomi released the flagship AI model MiMo-V2-Pro, which many perceived as an updated version of DeepSeek.

Xiaomi – a new star in the AI world

Recently a powerful model appeared on the OpenRouter platform that many mistakenly thought was a product of the DeepSeek startup. It turned out that it belongs to the Chinese giant Xiaomi. Earlier, the appearance of this model sparked speculation that DeepSeek was already testing its own “new” language system.

Why this matters
* DeepSeek‑V3 and V4 – their launch last year led to a drop in technology stocks worldwide. Investors began to doubt the need to spend billions on computing power for AI.

* Since then, the market has been eagerly awaiting the release of DeepSeek‑V4 to see what new achievements the Chinese startup will bring.

Where Hunter Alpha came from
At the beginning of the month, a free model called Hunter Alpha appeared on OpenRouter without a developer listed. Xiaomi, through its AI division and former DeepSeek engineer Luo Fuli (Luo Fuli), announced:

> *Hunter Alpha is an early internal test build of MiMo‑V2‑Pro*, the company’s flagship model optimized for agent interaction tasks.

The release of the new model coincided with the growing popularity of OpenClaw – a framework that allows creating AI agents capable of performing various actions on behalf of users. In China, OpenClaw is gaining momentum.

> “I call this a quiet ambush… We don’t even believe how quickly the paradigm has shifted from chats to agents,” Luo Fuli shared on X.

What MiMo‑V2‑Pro is
* Support for five key frameworks for building AI agents, including OpenClaw.
* Free trial access for one week.
* Xiaomi’s Hong Kong shares rose 5.8 % after the announcement.

MiMo‑V2‑Pro is the company’s “flagship base model,” intended as the brain of agent systems: coordinating complex processes, managing engineering tasks, and delivering reliable results.

In the global Artificial Analysis Intelligence ranking it placed 8th, and among Chinese language models it is second. The model contains over 1 trillion parameters (42 billion active) and a context window of 1 million tokens.

Performance
* In code generation it outperformed Claude 4.6 Sonnet.
* Overall performance on agent tasks is comparable to Opus 4.6.
* Stability and tool invocation accuracy have improved thanks to training optimizations.

In the future, Xiaomi plans to increase the efficiency of solving complex problems that require reasoning, long‑term planning, and systematic decision making.