Xiaomi AI: MiMo LLM Outperforms OpenAI, Lowers Costs

Xiaomi has significantly expanded its AI portfolio, now including advancements in language, reasoning, code, and voice technologies. The company first entered the open-source LLM race with MiMo, a 7 billion-parameter model. A variant of MiMo scored 95.8% on the MATH-500 benchmark, outperforming OpenAI's o1-mini and Alibaba's Qwen-32B-Preview on AIME 2024 and 2025. Xiaomi also released MiMo-V2-Flash, a larger model with 309 billion parameters, but only 15 billion active during inference. This model rivals top-tier models on software-engineering tests and offers an inference cost reportedly at 2.5% of Claude's. It achieves 150 tokens per second, with API input pricing at $0.1 per million input tokens. Additionally, Xiaomi introduced miclaw, an autonomous agent built on MiMo-V2-Pro, and OmniVoice, an open-source multilingual voice cloning model supporting over 600 languages. The bottom line is these developments show Xiaomi's broad push into advanced AI capabilities and open-source contributions.