Xiaomi has launched MiMo-V2-Pro, a large language model the Chinese electronics and car manufacturer claims can compete with leading American AI systems at significantly lower cost. Carl Franzen reports for VentureBeat that the model approaches the performance of OpenAI’s and Anthropic’s top offerings while costing roughly one sixth to one seventh as much per use.
The project is led by Fuli Luo, who previously worked on the influential DeepSeek R1 model. Luo described the release as a “quiet ambush” on the global AI frontier and indicated that an open-source variant is planned once the model reaches sufficient stability.
MiMo-V2-Pro contains one trillion parameters in total, though only 42 billion are active during any single operation. This keeps costs low without sacrificing capability. The model can process up to one million tokens of text at once, making it suitable for handling large volumes of enterprise data.
Independent benchmarking firm Artificial Analysis ranked MiMo-V2-Pro tenth on its global Intelligence Index, placing it alongside GPT-5.2 Codex and above Grok 4.20 Beta. On ClawEval, a test for autonomous task execution, it scored 61.5, close to Claude Opus 4.6 at 66.3.
Pricing starts at one dollar per million input tokens for queries under 256,000 tokens, compared to 1.75 dollars for GPT-5.2 and five dollars for Claude Opus 4.6. The model currently supports text only. Xiaomi has indicated a separate multimodal version is in development.
