Xiaomi is no longer moving like a company that only builds affordable phones and smart home devices. Over the past one and a half years, it has pushed harder into artificial intelligence and begun to look like a more serious global contender.
The shift is visible in a growing set of efforts that go beyond hardware, from large language models to voice cloning and even autonomous phone agents. Xiaomi now appears intent on shaping the direction of AI competition rather than simply following it.
MiMo-7B marks Xiaomi’s entry into large language models
The clearest sign of that ambition arrived in April 2025, when Xiaomi released MiMo-7B, its first open-source large language model. The name comes from Xiaomi Model, drawn from the “Mi” and “Mo” syllables.
Instead of positioning the model as a general chatbot, Xiaomi focused development on reasoning and coding. Those two areas require deeper logic than standard text conversation and give the model a more technical purpose.
That choice sets Xiaomi apart from many of its rivals. While other companies often emphasize broad conversational experiences, Xiaomi took a more specialized and functional route.
Small model, strong results
MiMo-7B has 7 billion parameters, yet Xiaomi says it performs beyond what its size might suggest. On the MATH-500 mathematics benchmark, the reinforcement learning version reportedly reached 95.8 percent.
The result becomes more notable because the model is also said to have surpassed OpenAI’s o1-mini and Alibaba’s Qwen-32B-Preview on the AIME mathematics benchmark for the 2024 and 2025 sessions. Those results place Xiaomi among the new names drawing attention in high-end AI competition.
Training scale helped shape the model
Xiaomi says MiMo-7B was trained with a specialized dataset containing 200 billion reasoning tokens. In total, the model used 25 trillion training tokens spread across three intensive training phases.
That scale suggests Xiaomi was not simply launching a small model for visibility. The company appears to have built a technical base intended to compete in a field where depth and training quality matter a great deal.
Open-source access broadens its reach
MiMo-7B was released under the MIT license, making it open for developers worldwide to access, study, and download through Hugging Face. That move gives the model a wider audience beyond Xiaomi’s own ecosystem.
In AI, open access often accelerates adoption, experimentation, and community-driven improvement. For Xiaomi, it also fits a broader ecosystem strategy that combines stronger hardware with smarter, more connected software.
AI ambitions extend beyond one model
MiMo-7B is likely only one piece of Xiaomi’s larger AI plan. The company is also developing precise voice cloning technology and preparing autonomous phone agents, signaling a push deeper into future technology layers.
If that direction continues, Xiaomi could keep reshaping its image from a budget-device maker into a more serious AI player. The company’s move into LLMs suggests it wants to be seen not just as a device manufacturer, but as a builder of a smarter digital ecosystem.
