In a blog post Thursday, Sundar Pichai and Demis Hassabis of Google have announced the launch of Gemini 1.0 Ultra, enhancing Google’s AI capabilities, and introduced its successor, Gemini 1.5, marking a significant advancement in AI technology. Gemini 1.5, led by the efforts of Google DeepMind, showcases dramatic improvements in efficiency and long-context understanding, able to process up to 1 million tokens, setting a new benchmark for large-scale foundation models.

Gemini 1.5 Pro, the first model of this new generation, is optimized for a wide range of tasks and offers performance comparable to the larger 1.0 Ultra model while utilizing less compute resources. It features a breakthrough in processing vast amounts of information, from extensive codebases to lengthy documents, and introduces an experimental feature allowing a context window of up to 1 million tokens, available in private preview through AI Studio and Vertex AI.

This model is built on a Mixture-of-Experts (MoE) architecture, enhancing its efficiency and ability to learn complex tasks quickly. It promises to revolutionize how AI models understand and process large datasets, offering new possibilities for developers and enterprises in creating and utilizing AI technologies.

Moreover, Gemini 1.5 has undergone rigorous safety and ethics testing in line with Google’s AI Principles, ensuring its deployment is responsible and aligned with global standards. The advancements in Gemini 1.5 represent a leap forward in AI, with Google aiming to responsibly bring these capabilities to a global audience, offering early access to developers and enterprises for testing and experimentation.

Featured image: Credit: Google