GitHub radar

MiMo-V2.6 by Xiaomi: a local AI agent for code and terminal tasks

This is a small model Xiaomi fine-tuned to handle tools and the terminal on its own. Install it for free via Ollama and get an agent for simple coding tasks right on your machine.

01XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B 68919k downloads/mo9B paramsimage-text-to-text

A 9-billion-parameter model (the internal settings a model uses to build its answers), built by Xiaomi on the open Qwen3.5-9B and further trained to use tools, the terminal, and write code on its own.

When it helps

Useful for routine coding tasks and experiments — free and without sending your code to someone else's cloud. For serious bugs and complex projects it's still weaker than larger models.

Pros

  • Runs fully locally, no subscription or payment needed
  • Much better at self-driven agent tasks (5→30 points on a key test)
  • Can read images, not just text
  • Your code never leaves your machine

Cons

  • Barely improved at fixing real bugs (SWE-bench: 60.0 → 61.1)
  • Weaker than large models like Opus for serious work

How to set it up — step by step

  1. 1Open your coding AI agent (like Claude Code) inside a test project folder.
  2. 2Send the agent the text from the block below.
  3. 3Wait for Ollama and the model to install — a regular 16GB laptop is enough.
  4. 4Give the model a simple test task in a separate folder, not your real project.
  5. 5Check how it uses tools and the terminal on its own before trusting it with anything important.

Text for your agent

Copy this and send it to your agent — Claude Code, Codex, any of them:

Install the MiMo-V2.6 model by Xiaomi through Ollama: ollama run mimo-v2.6 — a regular laptop with 16 GB of RAM is enough. Give it a simple task in a test repository and show me how it uses tools on its own.

A regular laptop with 16 GB of RAM or more — the default version of the model takes about 7 GB.

Open on Hugging Face