Xiaomi unveils MiMo, an open-source AI model rivaling OpenAI and Alibaba

Consumer tech giant Xiaomi has recently unveiled its very own open-source artificial intelligence (AI) model, MiMo. According to the China-based brand, the AI model was trained completely in-house, and even outperforms OpenAI’s o1-mini and Alibaba’s QwQ-32B-Preview models in math reasoning and coding.

The launch of MiMo aligns with earlier reports that Xiaomi was building up its computing resources, including procuring about 10,000 graphics processing units (GPUs) to train its models. Moreover, the brand had also made futile attempts to hire Luo Fuli, a DeepSeek employee who contributed heavily to the DeepSeek-V2 model.

Xiaomi’s venture into developing its proprietary large language model (LLM) stems from its goal to incorporate in-house AI technology into its products, such as EVs, smartphones, and home appliances.

According to Xiaomi, MiMo was trained using reinforcement learning by the brand’s specialized AI task force known as Core.

Source: [1] South China Morning Post

Read more of our articles below!