Official Launch of the Second Generation of K2 Think
End-to-End Open AI: Mohamed bin Zayed University Unveils the Second Generation of K2 Think
In a move that reflects the accelerating pace of innovation in foundational models, Mohamed bin Zayed University of Artificial Intelligence, in collaboration with G42 Group and Cerebras Systems, has announced the launch of the second version of the reasoning model "K2 Think", an advanced system specifically designed to handle complex tasks requiring logical thinking and multi-step reasoning.
This launch reaffirms the university's position as a global hub for developing open-source AI models, with a clear focus on scientific transparency, reproducibility, and suitability for research and industrial use.
From a Reasoning Model to an Integrated System
The new version of "K2 Think" is built on the foundational model K2-V2, which the university describes as the most powerful open-source advanced model developed by the Institute for Foundational Models to date. The system comprises approximately 70 billion parameters, giving it a high capacity to handle deep reasoning tasks across multiple domains.
The most notable shift in this version is the transition of "K2 Think" from being a reasoning model built on top of a foundational model, to a reasoning system integrated within the model's own architecture, which enhances its efficiency and grants it greater autonomy in handling complex tasks.
Full Transparency from Data to Evaluation
What distinguishes the second version is that it is presented as a fully open-source model from end to end. The university has made all stages of development available for inspection and reproduction, starting from pre-training data and its filtering mechanisms, through post-training recipes and reasoning alignment, all the way to evaluation methodologies and benchmark results.
According to the announcement, the model was trained using a fully independent data pipeline prepared and refined by the Institute for Foundational Models, including the "Guru" dataset, with any overlap with subsequent evaluation sets removed, in order to ensure a fair and reliable measurement of the model's performance.
Advanced Reasoning Capabilities and Long Context
"K2 Think" was specifically designed for complex reasoning tasks, processing problems step by step through long chains of thought, making it suitable for advanced applications in fields such as:
-
Advanced Mathematics
-
Science and Engineering
-
Programming
-
Logic and Simulation
The second version also provides an extended long context, allowing the model to retain information for longer periods and execute multi-stage reasoning operations across large volumes of knowledge content without losing coherence.
Strong Results in Reasoning Benchmarks
The university indicated that the second version of "K2 Think" achieved leading results when compared to other open-source reasoning systems, across a number of specialized benchmarks, including:
-
AIME2025
-
GPQA-Diamond
-
HMMT
-
IFBench
These indicators are considered an important benchmark for measuring the efficiency of models in logical reasoning and solving complex problems.
Access and Usage Methods
The developers have made several channels available for interacting with the model, including:
-
Direct web-based experience
-
Official website: k2think.ai
-
Smartphone applications on iOS and Android
-
Written explanatory materials and introductory visual content
A Global Research Ecosystem
It is worth noting that Mohamed bin Zayed University of Artificial Intelligence is a research university headquartered in Abu Dhabi, specializing in the development of AI technologies and advancing scientific research globally. The university launched its first bachelor's degree programs in artificial intelligence in 2025, with two tracks: Business and Engineering.
Cerebras Systems is a company specializing in AI computing infrastructure, known for developing the Wafer Scale Engine (WSE), which is used to run AI workloads within enterprises and via the cloud.
The Institute for Foundational Models, affiliated with the university, operates from centers in Abu Dhabi, Paris, and Silicon Valley, and leads the development of a number of open-source models, in addition to its contributions to building language models for low-resource languages, including the Arabic model Jais.
This article comes in the context of the growing technological transformation, drawing on multiple sources including Arabic websites and specialized blogs, alongside what is offered by Egyptian stores, Kuwaiti stores, vitamin stores, and foreign websites, while relying on Mashhor for social media services as a primary source of information and updates.
And with that, dear brothers and sisters, we have successfully completed the mission ✌
Send blessings upon the Prophet, and your hearts will be purified — do good no matter how small 🎯🌷
And do not forget our brothers and sisters everywhere in your prayers 📌
Greetings from the team #Ezznology #Ezz_AlTaqnia
You can also browse our store's products from here 👈#our store 🌷or here
To join our family on the Telegram group click 👈here
Also join our Facebook group where we share information and help members 👈here
Other topics that may interest you:
13 Transformations AI Is Leading in the Banking Sector in 2026
7 Fatal Mistakes Users Make When Connecting to Public Wi-Fi
Apple Redraws the Future of Siri: An Upcoming Update with Gemini Support to Enhance AI Capabilities
Thaura: An Arabic AI Model Redefining Privacy, Values, and Sustainability 2026
SIRBAI Unveils an Intelligent Platform for Managing Autonomous Drone Swarms
