Luo Fuli

Luo Fuli has moved from influential open-model research to leadership of Xiaomi’s MiMo programme. Her recent work spans efficient reasoning models, multimodal systems and embodied intelligence, placing her among the young technical leaders defining how China’s next generation of openly released foundation models will be built.

Luo Fuli leads Xiaomi’s MiMo large-language-model programme, after research roles that included contributions to the DeepSeek-V2 and DeepSeek-Coder-V2 papers. Those earlier credits matter, but they do not make her the sole creator of DeepSeek. Her present case is the programme she now directs. Luo appears in the Xiaomi LLM-Core Team author record for MiMo-V2-Flash, a 309-billion-parameter mixture-of-experts model with 15 billion active parameters. The January 2026 report described a hybrid attention design, pretraining on 27 trillion tokens, context extension to 256,000 tokens and multi-teacher on-policy distillation; it also reported up to a 2.6-times decoding speed-up through speculative decoding. She co-authored MiMo-Embodied, released in late 2025 and revised in 2026, which combined autonomous-driving and embodied-AI tasks in one open model. In April 2026, the team released MiMo-V2.5 weights under the MIT licence, including a full-modal model and an agent-focused variant with one-million-token context. These are very large team achievements. Luo’s distinction is technical direction across language, reasoning, multimodality and physical-world applications within a rapidly expanding open-model programme.