Xiaomi has made an entry into the frontier AI race with the launch of MiMo-V2-Pro — a one-trillion parameter foundation model whose performance benchmarks are said to be approaching those of OpenAI and Anthropic but at roughly one-sixth to one-seventh of the cost when accessed through its API. The project was led by Fuli Luo, who previously worked on DeepSeek R1 project — the Chinese AI model that shook the industry and cost Nvidia billions of dollars in market cap in a day. That model delivered frontier-level performance at relatively lower costs.
Luo did not mince words about what Xiaomi was attempting with this release, describing it in a post on X as “a quiet ambush” on the global AI frontier. Luo also confirmed that the company plans to open source a variant of MiMo-V2-Pro but “when the models are stable enough to deserve it.”
Xiaomi is the world’s third-largest smartphone manufacturer and has also forayed into the electric vehicle market with focus on combining hardware, software and now advanced AI reasoning under one roof. As per a report by VentureBeat, Xiaomi is trying focus on what it calls the “action space” of intelligence.
