Xiaomi has released its first open-source large model, “Xiaomi MiMo”.

1. Xiaomi’s First Large Language Model Release?

On April 30, 2025, Xiaomi—best known for its consumer electronics and electric vehicles—open-sourced its first reasoning-focused large language model (LLM), Xiaomi MiMo. The model is optimized from pre-training to post-training, significantly enhancing reasoning capabilities.

https://github.com/XiaomiMiMo/MiMo

https://huggingface.co/XiaomiMiMo

In benchmark tests:

• Mathematical Reasoning (AIME 24-25)

• Coding (LiveCodeBench v5) MiMo-7B, despite its compact ​7-billion-parameter​ size, outperformed:

• OpenAI’s proprietary o1-mini

2. How Capable is MiMo?

Xiaomi’s release focuses on dedicated reasoning performance. The RL-tuned MiMo-7B-RL excels in: ✔ ​Math problem-solving​ ✔ ​Code generation​ ✔ ​General logical reasoning​

2.0.1 Limitations:

• Small Scale (7B vs. Deepseek’s 671B): Lacks versatility in multimodal tasks, agent functions, and broad knowledge (e.g., humanities, arts, history).

• Analogy: Think of MiMo as a high school math whiz—strong in logic but not yet a polymath like encyclopedia-style models.

In the field of large language models, Xiaomi is a newcomer. will keep an eye on their progress.