By Brent Li
Chinese technology group Xiaomi launched its three latest large language models (LLMs) on March 19. The lineup includes MiMo-V2-Pro, a flagship model with over one trillion parameters designed for complex tasks; MiMo-V2-Omni, a multimodal model capable of simultaneously processing text, images, audio, and up to one hour of video; and MiMo-V2-TTS, a highly realistic voice synthesis model.
The capabilities of Xiaomi’s older LLMs lagged significantly behind the industry’s leaders. But these new models show the company is catching up, earning a chance to challenge top tier AI developers. The launch makes Xiaomi one of the few smartphone makers globally to possess highly capable, in-house LLMs — though the company has long maintained that it is much more than just a phone brand.
The DeepSeek connection
The new product suite was reportedly developed under the leadership of Luo Fuli, a prominent Chinese AI researcher who previously worked at Alibaba and later joined DeepSeek, where she was a key researcher working on the DeepSeek-V2 model. Chinese media have widely reported that Luo was personally recruited by Xiaomi founder and CEO Lei Jun and that she is being paid, in yuan terms, an eight-figure annual salary.
Marketing claims versus reality
In its promotion of the new models, Xiaomi has heavily emphasized their “agent” capabilities to act as autonomous assistants that understand complex workflows and interact with external applications. Tencent Tech noted that in past weeks, a blind test version of the model on the OpenRouter platform processed one trillion tokens.
The company claims the overall capability of the MiMo-V2-Pro is close to that of Anthropic‘s Claude 4.6 Opus, although in reality there is often a noticeable gap between marketing claims and real-world performance. With an overwhelming number of AI performance leaderboards to choose from, companies can easily cherry pick rankings that favor their own narrative. Based on The Insight Asia’ subjective, simple testing, MiMo’s core coding abilities are approaching industry standards. Yet, in complex reasoning and image understanding, it still trails mainstream leaders. Of course, such simple tests can only scratch the surface of an LLM’s full potential.
As agent frameworks like OpenClaw gain explode in global popularity, major AI companies are recognizing the massive revenue potential these frameworks create, as autonomous agents consume a high volume of tokens to complete tasks in their own complex way.
Xiaomi has now entered the fray. Following a familiar strategy among Chinese tech firms, the company claims its latest models mirror the performance of Claude Opus 4.6 but at a fraction of the cost. At just $1 per million input tokens, Xiaomi’s pricing is significantly lower than its U.S. competitors. Boldly raising the stakes, Lei Jun also asserted that Xiaomi’s new models outperform Elon Musk’s Grok in key metrics.
Transforming the smartphone OS
Given the MiMo models’ heavy emphasis on agentic capabilities, expectations are mounting that more powerful AI agents will be deeply integrated into Xiaomi’s smartphone operating systems. The industry is watching closely to see if Xiaomi can achieve a breakthrough in system-level automation, allowing phones to truly understand user intent and navigate apps autonomously.
The last pioneer to move aggressively into this space was ByteDance which partnered with smartphone maker ZTE to launch the Doubao smartphone assistant. However, because it directly simulated human screen taps to retrieve information or perform tasks from apps, it was blocked by many major app developers who viewed such behavior as a security threat. Since Xiaomi exerts deeper control over its operating system and its smartphone software ecosystem, its foundational models might give it a better chance to bring novel AI features to its users without the barriers that impeded its predecessors.
About two weeks before the launch of MiMo-V2, Xiaomi had already begun closed testing for Xiaomi Miclaw, a smartphone AI agent. By using an integrated, self-developed model, to power Miclaw, Xiaomi can achieve deeper system-level optimization and lower operational costs — key factors in scaling the technology across its massive global user base.
Sources:
