Xiaomi MiMo released MiMo-V2.6-Distill-Qwen-9B. It is a 9B parameter agentic model. It was fine-tuned from Qwen3.5-9B on MiMo-generated data. It covers coding, general agent tasks, visual coding and cybersecurity. The checkpoint is released for open research in agentic reinforcement learning. It outperforms Qwen3.5-9B on all reported benchmarks. Training used 77.4B tokens with 27.2B loss-bearing. It requires SGLang with Qwen3.5 support to run. The model is available on Hugging Face. It has been downloaded 13,669 times in the last month.
What changed
Qwen3.5-9B was the base model without MiMo-generated fine-tuning.
What it unlocks
Running a 9B agentic model fine-tuned for coding, general tasks, visual coding, and cybersecurity.
- 9B parameters
- 77.4B total training tokens
- 27.2B loss-bearing tokens
- 13,669 downloads last month
What you need to act on it
- SGLang with Qwen3.5 support
- local deployment or self-hosting
Sources