"The models have improved tool-use and reasoning capabilities, understand image and text inputs, are faster and more efficient, and are designed to support 15 languages.” ...