Skip to content

How-ToRAM

How to Run Xiaomi's MiMo-V2.6 Coding Model Locally With Ollama

Xiaomi's MiMo-V2.6-Distill-Qwen-9B is an MIT-licensed 9B coding and agent model with image input and a 256K context window. Run it with `ollama run mimo-v2.6`: the 4-bit build is 6.6GB and suits 8GB-plus graphics cards, while the 8-bit build is 10GB. Raise the context length for agent work. Xiaomi's benchmark gains are its own figures. Forge PC has not tested MiMo-V2.6.

By , Editorial team

3 min read

How to Run Xiaomi's MiMo-V2.6 Coding Model Locally With Ollama

Xiaomi's MiMo team has released MiMo-V2.6-Distill-Qwen-9B, a small open model aimed at coding and agent workflows. Ollama's library page already shows more than 96,000 downloads, and at 9 billion parameters it's small enough to run on a mainstream gaming graphics card.

Forge PC has not tested MiMo-V2.6 or measured its speed. This guide is based on Ollama's model page and tag list and Xiaomi's Hugging Face model card, checked on 16 November 2026. Benchmark results quoted below are Xiaomi's own.

What MiMo-V2.6 is

According to Ollama's page, MiMo-V2.6-Distill-Qwen-9B was fine-tuned from Alibaba's Qwen3.5-9B on data generated by Xiaomi's larger MiMo models. It:

  • works with text and images, so you can give it screenshots or interface mock-ups;

  • supports thinking (step-by-step reasoning) and tool use, which coding agents need;

  • supports up to 256K tokens of context;

  • is released under the MIT licence, which allows commercial use.

Xiaomi pitches it for understanding unfamiliar code, investigating failures and drafting fixes, multi-step agent tasks, "visual coding" from screenshots and explaining potential security vulnerabilities.

Xiaomi's benchmark claims

Xiaomi reports these results against the Qwen3.5-9B model it was built from:

Benchmark

Qwen3.5-9B

MiMo-V2.6-Distill-Qwen-9B

SWE Verified (avg@3)

60.0

61.1

SWE Pro (avg@3)

32.0

44.6

Terminal Bench 2.1 (avg@1)

27.0

37.1

Toolathlon-Verified (avg@1)

25.9

35.2

These are the developer's figures, not independent tests. They suggest the biggest gains are in agent-style tasks that involve tools and the terminal, rather than a big leap in raw coding ability. Treat them as a reason to try the model, not proof it suits your work.

Which version fits your graphics card?

Ollama offers several builds. The main ones for Windows PCs are:

Ollama tag

Download size

Notes

mimo-v2.6 (default, same as 9b)

6.6GB to 8.7GB

Ollama lists this size range for the default build

mimo-v2.6:9b-q4_K_M

6.6GB

4-bit quantisation

mimo-v2.6:9b-q8_0

10GB

8-bit, closer to full quality

mimo-v2.6:9b-gguf-bf16

19GB

Full precision

There are also MLX builds (mxfp8, nvfp4 and mlx-bf16), which are for Apple Silicon Macs rather than Windows PCs.

As a rough guide, the 4-bit build should suit graphics cards with 8GB or more of video memory, the 8-bit build is better suited to 12GB or 16GB cards, and the full-precision version needs a 24GB card or more. Remember that the context window uses memory too, so leave headroom. Our guides to running LLMs on 8GB of VRAM and GGUF quantisation explain the trade-offs.

How to run it

  1. Install or update Ollama from ollama.com. New models sometimes need a recent version.

  2. Open Terminal and run ollama run mimo-v2.6. Ollama downloads the default build and opens a chat. To choose a specific build, use its tag, for example ollama run mimo-v2.6:9b-q8_0.

  3. Check it's running on your GPU with ollama ps. If the PROCESSOR column shows part of the model on the CPU, try a smaller build or a shorter context.

  4. Give it images in the Ollama app, or see our guide to vision models in Ollama.

If you'd prefer a ChatGPT-style interface in your browser, follow our Open WebUI guide.

Set a longer context for coding

Ollama's default context window is only 4k tokens on PCs with less than 24GiB of video memory, and it recommends at least 64,000 tokens for coding tools and agents. You can raise it with the context slider in the Ollama app's settings. A 64k context uses considerably more memory than 4k, so check ollama ps again afterwards. Our Ollama speed guide has more tips.

Use it with coding tools

Ollama's page lists ready-made commands for several agents, including:

ollama launch claude --model mimo-v2.6
ollama launch opencode --model mimo-v2.6

You can also use it in your editor through Continue, in the terminal with Aider, or in any app that supports Ollama's OpenAI-compatible API.

Our verdict

MiMo-V2.6 is worth a try if you want a small, permissively licensed coding model that can also read screenshots and use tools. The 4-bit build fits mainstream 8GB-plus graphics cards, and it plugs straight into Claude Code and other agents through Ollama. Just remember that the benchmark gains are Xiaomi's own, a 9B model has clear limits on big projects, and you'll need to raise Ollama's context length for agent work. Forge PC has not tested MiMo-V2.6.

The bottom line

MiMo-V2.6 is worth a try if you want a small, permissively licensed coding model that can also read screenshots and use tools. The 4-bit build fits mainstream 8GB-plus graphics cards, and it plugs straight into Claude Code and other agents through Ollama. Just remember that the benchmark gains are Xiaomi's own, a 9B model has clear limits on big projects, and you'll need to raise Ollama's context length for agent work. Forge PC has not tested MiMo-V2.6.

Sources

  1. 01Ollama, mimo-v2.6 model library page
  2. 02Ollama, mimo-v2.6 tags
  3. 03Xiaomi MiMo, MiMo-V2.6-Distill-Qwen-9B model card
  4. 04Ollama, Context length

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

How to Enable AMD Fluid Motion Frames (AFMF 2.1) on a Radeon Graphics Card

How-To

How to Enable AMD Fluid Motion Frames (AFMF 2.1) on a Radeon Graphics Card

To enable AMD Fluid Motion Frames, install Adrenalin 25.3.1 or newer, then select the HYPR-RX profile (RX 7000 and newer) or turn on AMD Fluid Motion Frames under Gaming > Graphics, globally or per game. AFMF 2.1 works on RX 6000 cards and newer in DirectX 11, DirectX 12, Vulkan and OpenGL games, with VSync off; RX 6000 cards need exclusive full screen.

Forge PC Staff2 min

Two vs Four RAM Sticks for DDR5: Why 2x32GB Usually Beats 4x16GB

Versus

Two vs Four RAM Sticks for DDR5: Why 2x32GB Usually Beats 4x16GB

Two DDR5 sticks are better for most builds. AMD lists DDR5-5600 as the AM5 maximum with two modules but DDR5-3600 with four, and XMP or EXPO speeds above that are overclocking and not guaranteed. Buy 2x16GB or 2x32GB for 32GB or 64GB, and use four sticks, ideally a matched kit, only if you need 128GB or more.

Forge PC Staff3 min

AMD's FSR 4.1.2 Unreal Engine 5 Plugin Brings ML Upscaling to RX 6000 Cards

News

AMD's FSR 4.1.2 Unreal Engine 5 Plugin Brings ML Upscaling to RX 6000 Cards

AMD's FSR Unreal Engine 5 plugin now uses FSR Upscaling 4.1.2, bringing ML-based upscaling to Radeon RX 6000 cards and select Ryzen integrated graphics as well as RX 7000 and RX 9000, in UE 5.3 to 5.8 games that adopt it. It's DirectX 12 only for now and adds Frame Generation 4.0.2 with driver-assisted frame pacing on RX 9000 cards.

Forge PC Staff2 min

NVIDIA's DGX Spark HPC Developer Kit Links Four Sparks Into a 512GB Desk-Side Cluster

News

NVIDIA's DGX Spark HPC Developer Kit Links Four Sparks Into a 512GB Desk-Side Cluster

NVIDIA's DGX Spark HPC Developer Kit, announced at SC26 on 16 November 2026, links four DGX Spark systems in a ring over ConnectX-7 with no network switch, for 512GB of combined memory and 20TB of NVMe storage. NVIDIA says it can spread models of up to 700B parameters across the nodes with NVFP4. No UK kit price yet; Scan sells two-Spark bundles at about £12,845.

Forge PC Staff3 min