On March 31, Lei Jun, founder, chairman, and CEO of Xiaomi, shared the latest "achievements" of the large model MiMo-V2-Pro on a social platform. In the authoritative evaluation list Text Arena, known as the "large model college entrance exam," MiMo-V2-Pro successfully ranked among the top five globally in the Model Rank (model ranking) category, thanks to its outstanding performance in complex logical reasoning, long instruction following, and multi-turn dialogue.

image.png

Rejecting "ranking chasing": The fifth globally voted by real users

This breakthrough of Xiaomi MiMo-V2-Pro carries high value, as Text Arena features a unique "double-blind test" mechanism:

  • Model anonymity: The model's identity is completely hidden during the evaluation process.

  • Instant voting: Global real users conduct blind tests and vote instantly based on the quality of the answers.

  • Eliminating bias:Lei Jun emphasized that this mechanism completely eliminates common "data set ranking chasing" behaviors seen in traditional evaluations. Currently, Xiaomi ranks just behind Anthropic, OpenAI, and Google on this list.

R&D strength: Ranking among the global top tier in multiple dimensions

Aside from the single model ranking, Xiaomi also showed strong performance in the rankings measuring the comprehensive R&D capabilities of the laboratory:

  • Laboratory ranking: In the ArenaExpert (expert evaluation) dimension, Xiaomi ranks fourth globally.

  • Code ability: In the Code Arena dimension, Xiaomi ranks fifth globally.

  • Call popularity: Since its release on March 19, the flagship model MiMo-V2-Pro has maintained the top position in the daily, weekly, and trend charts of the global call volume list OpenRouter.

Strategic investment: 6 billion yuan over three years, AI-driven full ecosystem

This achievement reflects Xiaomi's substantial investment in the AI field. Lei Jun previously stated that Xiaomi will actively embrace the AI era and invest 6 billion yuan in the AI field over the next three years. From the AI transformation of the HyperOS system-level input method to the full opening of the Xiaomi MiMo agent framework, Xiaomi is striving to deeply integrate the capabilities of large models into the "people, car, home, and all ecosystem."

Time-limited offer: Official announcement of extended free period

To allow more developers to experience the peak power of domestic large models, Xiaomi MiMo officially announced that the "first week free trial of the joint global top Agent framework" originally scheduled to end at the end of March will be extended to two weeks. The free period will continue until 24:00 on April 2, 2026.

Conclusion: From "score leading" to "experience leading"

When Xiaomi's self-developed large model starts competing with top-tier giants in international anonymous evaluations and wins, it marks that domestic large models have already gained extremely strong global competitiveness. As Lei Jun said, Xiaomi's core business is making smart devices, and a powerful AI foundation will become the most solid moat for Xiaomi's hardware products in the second half of the competition.