Kimi K3.1 Pro Open Weights: What Moonshot AI Just Released

A 2.8T-Parameter Model You Can Actually Download

We'll keep advancing open intelligence.

On November 4, 2026, Moonshot AI released the model weights for Kimi K3.1 Pro, publishing the source code and instructions in the official repository on huggingface.co. The model card lists it as an image-text-to-text model with 2.8 trillion parameters, making it one of the largest openly downloadable multimodal models to date.

The release follows the company's stated commitment to open intelligence, and the repository includes evaluation results alongside the weights themselves. The model tree, usage instructions, and license terms are all documented in the same place, so you do not need to hunt across multiple sources to get started.

What It Does and How You Can Use It

Kimi K3.1 Pro accepts both images and text as input and produces text output. That makes it suited to tasks like visual question answering, document understanding, captioning, and multimodal reasoning, where you feed in a screenshot, chart, or photo alongside a prompt and get a written answer back.

Because the weights are open, you can download them from the official repository on huggingface.co and run the model on your own infrastructure. The model card includes deployment and availability notes, so check those first to understand the hardware requirements before you commit to a local setup.

• Multimodal input

  Handles images and text together, so you can ask questions about visual content directly.

• Text output

  Generates written responses, making it useful for analysis, summarization, and Q&A.

• Open weights

  Downloadable from the official repository, so you control where and how it runs.

Access, Licensing, and What to Check Before You Commit

The model is available through the official repository on huggingface.co, where you will find the weights, source code, and step-by-step instructions. The repository also lists a license section, so review the terms carefully if you plan to use the model commercially or redistribute it.

At 2.8 trillion parameters, Kimi K3.1 Pro is not a lightweight download. Expect significant storage and compute requirements, and plan your deployment around the hardware you actually have. The model card's evaluation results give you a sense of its capabilities, but running it yourself is the only way to know how it performs on your specific tasks.