Chinese electronics and automotive manufacturer Xiaomi released MiMo-V2-Pro, a new 1-trillion parameter foundation model with benchmarks approaching those of OpenAI and Anthropic's flagship models. The company claims the model delivers comparable performance at around one-sixth to one-seventh the cost when accessed through proprietary APIs.
The project is led by Fuli Luo, a veteran of the DeepSeek R1 project, who characterized the release as a "quiet ambush" on the global AI frontier. Luo stated on X that Xiaomi plans to open source a model variant when it becomes sufficiently stable. The model uses a sparse architecture designed to maintain high-fidelity reasoning over massive data spans while minimizing latency and cost.
Xiaomi's entry into frontier AI builds on its established position as the world's third-largest smartphone manufacturer and recent expansion into electric vehicles. The company's hardware expertise in IoT devices and EVs like the SU7 and YU7 SUV informs MiMo-V2-Pro's architecture, which is designed to serve as the "brain" for complex systems ranging from supply chain management to autonomous coding agents.
This release signals intensifying competition in the large language model space, particularly from Chinese companies challenging U.S. AI leadership. Xiaomi's focus on the "action space" of intelligence and autonomous operation of digital systems suggests a shift beyond conversational AI paradigms toward more practical applications in physical-world engineering and system management.