MiMo-V2.6-Distill-Qwen-9B
Xiaomi MiMo-V2.6-Distill-Qwen-9B is an open-weights 9B vision-language reasoning model. It answers questions about images and text, with optional step-by-step thinking.
Overview
MiMo-V2.6-Distill-Qwen-9B is an open-weights vision-language model from Xiaomi's MiMo team. It distills the reasoning ability of the larger MiMo-V2.6 models into a 9B Qwen-based model. You give it a written instruction, optionally with one or more images, and it returns a text answer. With thinking enabled, the model first reasons step by step and then writes its final answer.
What you can build
- Image Q&A: ask questions about photos, screenshots, charts and diagrams
- Detailed image descriptions and alt text
- Step-by-step solutions for math, logic and science problems
- Coding help, such as writing functions, explaining errors and reviewing snippets
- Multi-image comparisons, such as spotting differences between two versions of a design
Inputs
- A written instruction in plain text. This is the only required input.
- Optional images in JPG, PNG, WebP or BMP format, up to 5 per run.
- Optional system instructions that set the model's tone, role or rules.
- A switch for thinking mode. It is on by default.
- Optional generation controls: maximum output length, temperature, top-k, top-p, repetition penalty and seed.
Outputs
The model returns plain text. When thinking is enabled, the reasoning appears first inside think tags, followed by the final answer. When thinking is disabled, only the answer is returned.
Recommended settings
- Keep the default sampling: temperature 0.6, top-k 20, top-p 0.95.
- Leave thinking on for math, logic and coding. Turn it off for simple descriptions and short answers.
- Thinking can be long. Keep the maximum output length at 8,192 tokens or higher for hard problems, or the answer may be cut off.
Limitations
- The output is text only. The model cannot generate or edit images.
- If the output length limit is reached during thinking, the run may end before the final answer is written.
- On some hard prompts the thinking can repeat itself until it reaches the length limit.
- Very small text, blurry photos and heavily compressed screenshots reduce accuracy.
- As a 9B model, it can make factual mistakes. Check important answers.
Safety & compliance
Review outputs before using them in medical, legal or financial decisions. Do not use the model to produce harmful content or to analyze images of people without their consent.
API quick start
Run MiMo-V2.6-Distill-Qwen-9B with a single API call.
{
"prompt": "Describe this image in detail.",
"inputImage": "https://your-cdn.com/input.png",
"system_prompt": "...",
"enableThinking": true
}curl -X POST "https://api.wiro.ai/v1/Run/xiaomimimo/mimo-v2-6-distill-qwen-9b" \
-H "Content-Type: application/json" \
-H "x-api-key: YOUR_WIRO_API_KEY" \
--data-binary @- <<'JSON'
{
"prompt": "Describe this image in detail.",
"inputImage": "https://your-cdn.com/input.png",
"system_prompt": "...",
"enableThinking": true
}
JSONOther models you might like
Explore similar image generation models.
