Xiaomi Releases Open-Weights MiMo-V2.6-Distill-Qwen-9B Agentic Model
Xiaomi has officially released MiMo-V2.6-Distill-Qwen-9B, an open-weights 9-billion parameter agentic language model fine-tuned from Qwen3.5-9B. The model was trained using supervised fine-tuning on synthetic datasets generated by Xiaomi's flagship MiMo AI models. This release provides the open-source AI community with a compact model tailored specifically for task execution and agentic workflows. It highlights the growing trend of using distillation from larger proprietary models to boost the performance of smaller, locally runnable LLMs. Built on the Qwen3.5 architecture, the model is designed for function calling, tool use, and multi-step reasoning. The open weights have been published directly to Hugging Face by Xiaomi's MiMo team for local deployment.
## BACKGROUND
Model distillation involves training a smaller language model using synthetic outputs generated by a larger, more powerful model to achieve high performance at lower computational costs. Agentic language models are specialized LLMs engineered to plan actions, call external tools, and complete multi-step autonomous workflows rather than just generate conversational text.