Ten months to Xiaomi's highest professional grade
Luo Fuli, the former DeepSeek researcher leading Xiaomi's MiMo artificial intelligence team, has reportedly reached level 22, the highest grade in the company's professional ranking system, just 10 months after joining. Her promotion appeared in an internal announcement on September 22, placing the 31-year-old researcher at the top of that career ladder as Xiaomi expands its investment in the models behind its AI products.
Contents
Luo confirmed her move to Xiaomi in November 2025. Her rise has since involved more than a new title: her team has gained responsibility for the underlying models used in XiaoAI, Xiaomi's virtual assistant, and pursued projects far larger than the company's first MiMo release.
Xiaomi's business gives those decisions a practical setting. The company sells smartphones, electric vehicles and other connected devices. Its AI researchers are developing technology that could serve products across that range, while separate teams handle the computing infrastructure and the work of putting AI into devices.
The promotion also brings renewed attention to an unconfirmed recruitment package. Founder and CEO Lei Jun reportedly sought to hire Luo in late 2024 with an annual salary of 10 million yuan; another description put the proposed salary at "tens of millions" of yuan. Neither Xiaomi nor Luo has confirmed the offer, and the promotion announcement disclosed no new compensation figure.
What Luo now oversees
MiMo is Xiaomi's family of large language models, the software systems trained to understand and generate text, reason through problems and carry out complex tasks. Luo leads the research team developing those models, which provide the underlying capabilities that other teams can build into services.
A July reorganization of XiaoAI expanded that responsibility. Under the new arrangement, Luo's team took charge of developing the foundational AI models. Other teams retained responsibility for cloud infrastructure and for integrating AI into phones, vehicles and other devices.
The division separates three connected jobs. Researchers build and improve the model. Infrastructure teams supply the computing systems needed to run it. Product teams connect it to features that customers can use. A better model can give a product more capable reasoning or language skills, while the final experience also depends on the other two jobs.
Luo described a broader ambition when announcing her move in a WeChat post. The researcher, who was joining Xiaomi's MiMo team, wrote:
Intelligence will step beyond language into the physical world.
Phones and vehicles provide Xiaomi with places to pursue that ambition. Luo's public statement outlined a direction for the research, while the organizational changes gave her team a defined role in supplying its technical foundation.
From a small model to much larger bets
Xiaomi released its first MiMo model in April 2025. Called MiMo-7B, it had 7 billion parameters, the numerical values adjusted during training that help determine how a model processes information and produces answers.
The model was developed under Xiaomi's Core AI task force using reinforcement learning. In this kind of training, a system receives feedback that rewards more successful responses, helping it improve at tasks such as solving problems.
Xiaomi said MiMo-7B outperformed OpenAI's GPT-4o and Alibaba Cloud's QwQ-32B on certain benchmark tests despite having fewer parameters. Benchmarks are standardized exercises used to compare systems. Performance on them gives researchers a way to measure particular skills, such as mathematical reasoning or coding.
Luo and her team subsequently chose to train two much larger models at the same time: one with several hundred billion parameters and another with 1 trillion. Those projects became MiMo-V2-Flash and MiMo-V2-Pro, according to reporting on the team's development decisions.
A trillion parameters is about 143 times the 7 billion in the original MiMo model. Parameter counts describe a model's scale; training methods, data and design also affect its capabilities. For Xiaomi, the move represented a substantial expansion in the size of the systems its researchers were trying to build.
The computing costs behind the research
Luo's team reportedly livestreamed a training dashboard in September 2026 showing a combined cost rate of US$30,800 per hour for two models undergoing reinforcement learning. At that displayed rate, 24 hours of continuous operation would cost US$739,200.
The dashboard figure describes the rate for those training activities. A total bill would depend on how long the systems ran and how spending changed during the work. The duration needed to calculate that total was not disclosed.
MiMo's expansion began before Luo publicly confirmed her arrival. The team reportedly had fewer than 20 people in January 2025 and around 50 to 60 by May or June. Xiaomi therefore had an existing research operation and a released model before her November announcement.
Lei reportedly received updates on MiMo about every two weeks and generally left technical choices to Luo and the researchers. That arrangement gave the team considerable room to decide how to pursue its larger models, including the choice to train two ambitious systems simultaneously.
The combination of larger models, substantial computing costs and access to the chief executive helps explain why Luo's position matters beyond its numerical rank. Her team's choices affect how Xiaomi commits research resources and which capabilities its product teams may eventually have available.
How Luo reached Xiaomi
Luo's career connects university language research, a major technology company's laboratory and the investment firm behind DeepSeek. Born in 1995 into a rural family in Sichuan province, she studied computer science at Beijing Normal University before earning a master's degree in computational linguistics at Peking University.
Computational linguistics uses computing to study and process human language. It connects closely with the work behind systems that interpret questions, translate text and generate written responses.
Her route through research and industry includes several key dates:
- 2019: Luo contributed to eight papers presented at the annual meeting of the Association for Computational Linguistics.
- After graduation: She joined Alibaba's DAMO Academy, its research institute.
- 2022: She joined High-Flyer Quant, the investment firm whose founders also established DeepSeek.
- May 2024: DeepSeek released DeepSeek-V2, with Luo among its contributors.
- April 2025: Xiaomi released MiMo-7B, before Luo publicly confirmed her move.
- November 2025: Luo announced that she had joined Xiaomi's MiMo team.
- September 22: Xiaomi reportedly announced her promotion internally, about 10 months after she joined.
At DeepSeek, Luo worked on DeepSeek-V2 and, according to reporting on her later career, DeepSeek-V3. She brought experience with models containing roughly 600 billion to 700 billion parameters to Xiaomi's effort to develop much larger systems.
DeepSeek-V2, released as an open source model in 2024, helped establish the Chinese company as a challenger to leading American AI developers. Open source releases allow people outside the developer's organization to access and work with model resources under the applicable license.
Why the recruitment drew attention
Lei's reported approach to Luo in late 2024 put a large salary figure at the center of public interest in her career. An annual package of 10 million yuan would equal roughly 833,333 yuan a month before tax, if that reported figure reflected an actual agreement.
The confirmed public milestone came later, when Luo announced in November 2025 that she had joined MiMo. The gap between the recruitment reports and her announcement left months of speculation about where she would work next.
Xiaomi's shares rose 1.7 percent to HK$43.70 at the close of trading on the Wednesday she announced her move. The recruitment gave the company a prominent researcher with direct experience developing models at DeepSeek, alongside its existing MiMo team.
The reported salary and the subsequent promotion concern different parts of that story. The salary describes an unconfirmed recruitment offer. Level 22 describes her reported standing within Xiaomi's professional ranking system. Her expanding responsibilities, meanwhile, can be seen in the assignment of foundational model development to her team and its pursuit of larger models.
For Xiaomi, hiring Luo brought experience relevant to decisions that are expensive to reverse once large training runs begin. Model design, training methods and the allocation of computing resources all shape how a research program develops.
Persistence behind the public profile
Luo has repeatedly rejected the description of her as an "AI genius girl." In a profile published by Peking University, she credited curiosity, execution and resilience for her progress.
Explaining her achievements in that university profile, Luo said:
I'm not a genius.
She described the qualities she believed had helped her reach that point:
Curiosity about fundamental problems, solid execution, and a bit of stubborn resilience, that's what brought me here.
Luo recalled struggling with computing during her early studies because she had little previous experience with computers. Problems that classmates solved in three minutes could take her three hours, a difference of 60 times in the time spent on an exercise.
Within two years, sustained work helped her become one of the strongest students in her class. Her contribution to eight papers presented at a major language research conference in 2019 later brought wider recognition.
Her account places the rapid promotion at Xiaomi alongside a much longer period of learning. The 10 months describe her reported climb within one company's ranking system; the research experience behind it stretches across university study, Alibaba, High-Flyer and DeepSeek.
A fast promotion, a longer ambition
Luo's public comments about her career favor steady progress over speed. In an interview with Chinese video creator Liang Zi, she used a modest comparison to explain how she viewed her development.
Describing her approach to career growth, Luo said:
I am like a snail who does not pursue the fastest move, but my every step is clearly visible.
Luo, whose husband also works in technology, has said she wants to continue advancing professionally while maintaining a happy family life. Those personal goals sit alongside a role that now carries responsibility for increasingly ambitious research.
For Xiaomi customers, the concrete work lies in connecting that research to products. Luo's team develops the underlying models; other teams must run them reliably and integrate their capabilities into phones, cars and connected devices. Her promotion recognizes her position inside that effort, while the broader program requires coordination across those separate responsibilities.
The next stage of MiMo's development will involve both the larger models and the systems built around them. Xiaomi's early MiMo release, the XiaoAI reorganization and the reported spending on training show how quickly the company has expanded the effort since April 2025.
Key Points
- Luo Fuli reportedly reached Xiaomi's highest professional grade, level 22, about 10 months after joining.
- Her promotion was announced internally on September 22.
- She leads MiMo, and her team gained responsibility for the foundational models behind XiaoAI.
- Xiaomi launched MiMo-7B in April 2025; later projects pursued models with hundreds of billions and 1 trillion parameters.
- A September 2026 training dashboard reportedly displayed combined costs of US$30,800 per hour for two models.
- Neither Xiaomi nor Luo has confirmed the reported recruitment salary.
- Luo credits persistence, curiosity and sustained work for her career progress.
