Meta Description: MBZUAI’s K2 Horizon AI models bring six openly shared AI tools to researchers everywhere. Here’s what makes this release worth appreciating.

K2 Horizon AI Models: MBZUAI’s New Open AI Release

Sharing knowledge openly is one of the kindest things a research institution can do, and Abu Dhabi’s Mohamed bin Zayed University of Artificial Intelligence, known as MBZUAI, has just done exactly that. As an AI university, it has released the K2 Horizon AI models, a set of six AI tools: 0.9B, 3.7B, 7B, 32B, 36B-A4B, and 375B-A23B parameters. Built to be openly available for researchers, developers, and institutions around the world. It is a warm and welcome move at a time when so much AI technology stays locked behind closed doors.

What Are the K2 Horizon AI Models?

The K2 Horizon AI models were developed at MBZUAI’s Institute of Foundational Models. Rather than releasing a single tool, the university built an entire family of models to suit different needs:

• Six models in total, ranging from 0.9 billion to 375 billion parameters
• The smallest model, at 0.9 billion parameters, is light enough to run on a smartwatch
• The 375B-A23B model is designed for demanding enterprise deployments, while the 32B and 36B-A4B models provide options for local hosting and on-premises environments. 
• The 0.9B model is designed for highly constrained devices such as watches and glasses, while the 3.7B and 7B models target phones and other on-device applications. 
• Every model ships with its weights, code, training data, and methodology included

This range means the K2 Horizon AI models can support everything from small, lightweight devices to large, demanding research projects.

Why Openness Actually Matters

This release stands out because it’s fully open AI. According to MBZUAI, this fully open AI approach allows other researchers, developers, and institutions to inspect and build upon the code, training data, and methodology behind each model. Nothing is hidden behind an API or a paywall.

Prof Eric Xing, the president of MBZUAI, explained that meaningful progress in AI depends on the ability to examine and improve technology, rather than simply access it. He described the release as a belief that the most important technology of our time should be built with the world, not kept from it. That is a genuinely generous outlook, and it deserves real appreciation.

Smart Features Built Into the Models

The K2 Horizon AI models include a few thoughtful technical touches that make them more practical to use:

• Dynamic model routing, which automatically directs tasks to the most cost-effective model available
• A mixture-of-experts architecture, designed to sharpen reasoning without needing extra computing power
• Compatibility with popular platforms like Hugging Face, Ollama, Unsloth, and the IFM Repository

Hector Liu, director of the Institute of Foundational Models’ Silicon Valley research lab, shared that developers can start with the smallest model, scale up to the flagship version, and verify every claim made about the models along the way. That kind of transparency builds real trust.

A Reflection of the UAE’s Wider AI Ambitions

This release is part of a bigger, encouraging pattern. MBZUAI, which opened in 2020 as the world’s first AI-dedicated university, continues to support the UAE’s broader push to grow its AI ecosystem as the country diversifies its economy. The country has built meaningful partnerships with companies like Microsoft, Nvidia, and OpenAI, and is also working with the US on a planned AI campus in Abu Dhabi that will include significant data centre capacity.

The K2 Horizon AI models fit naturally into this ongoing effort, adding openly available research tools into a growing regional AI landscape.

Why This Openness Deserves Appreciation

• It gives researchers everywhere a chance to study and improve real AI systems, not just use them
• It supports smaller developers and institutions who may not have access to closed, expensive AI systems
• It builds trust through open-weight models that can be checked and verified by the community
• It reflects a genuine belief in shared progress rather than closed competition

This kind of openness quietly benefits the wider AI university community, from students to independent researchers, by giving everyone a fair chance to learn from real, working systems.

Conclusion

K2 Horizon represents a significant contribution to the open AI ecosystem, giving researchers and developers access to models spanning edge devices through enterprise-scale deployments while providing greater visibility into how the models were developed. 

If you want guidance on how open AI tools like these could support your own projects or organisation, Horizontal8 is here to help. Reach out to Horizontal8 today, and let’s explore what these tools could mean for you.

Frequently Asked Questions

What are the K2 Horizon AI models? 

They are a set of six AI models developed by MBZUAI’s Institute of Foundational Models, ranging from 0.9 billion to 375 billion parameters, all released as fully open tools.

What does “fully open” mean for these models? 

It means the code, training data, and methodology behind each model are openly shared, allowing anyone to inspect, use, and build on them.

Can the smallest K2 Horizon model really run on a smartwatch? 

Yes, the smallest model, at 0.9 billion parameters, is small enough to run on wearable devices like smartwatches.

Where can these models be downloaded? 

K2 Horizon is released as a fully open model family, with weights, code, training information, and methodologies made available. The models and code are released under Apache 2.0, while datasets follow their applicable licenses. 

Who created the K2 Horizon AI models? 

They were developed by MBZUAI, the Abu Dhabi-based Mohamed bin Zayed University of Artificial Intelligence, through its Institute of Foundational Models.