ABU DHABI: An Abu Dhabi-based artificial intelligence research institute has released six AI models, along with their training data, code, and development methods, taking an unusually open approach as leading technology companies increasingly guard the details behind their most advanced systems.
The Institute of Foundation Models (IFM) unveiled the models on Thursday as part of its K2 Horizon release, providing researchers with access not only to the finished systems but also to the materials used to create them.
The release includes model weights, training datasets, code, methodologies and intermediate checkpoints, allowing independent researchers to examine different stages of development and attempt to reproduce the institute’s results.
IFM founder Eric Xing said the approach was intended to make the development process itself accessible, rather than merely releasing a finished model.
The initiative stands in contrast to a growing tendency among major AI developers to disclose relatively little about the datasets, training procedures and technical decisions behind their most capable models.
While a number of companies describe some of their AI systems as “open,” access is often limited to model weights or subject to restrictions. Providing training data and intermediate checkpoints gives researchers substantially greater ability to study how a model was built and how its behaviour developed during training.
The release also highlights the UAE’s expanding ambitions in artificial intelligence, as Abu Dhabi invests heavily in research infrastructure and seeks to establish itself as an international centre for advanced AI development.