Artificial Analysis ranks K2 think v2 highest for openness

Artificial Analysis has published an Openness Index to evaluate how open large language models are. The index assesses models based on the availability of training data and training regimens.

• K2 think v2 received the highest rating because it provides its training data and regimen, allowing recreation by those with sufficient resources.
• Deep Seek received a lower score as it does not publish its training methods or data.
• The evaluation uses an analogy comparing software binaries to source code availability.

The update highlights that providing full transparency enables others to recreate the model, which is considered a higher level of openness than simply supplying a free binary.