Google has announced its latest and most powerful AI model, Gemini, with state-of-the-art performance in the integration of text, code, audio, images and videos, among other areas.

The unveiling came from Sundar Pichai, CEO of Google and Alphabet; and Demis Hassabis, CEO and co-founder of Google DeepMind, via a release on Wednesday, December 6.

Hassabis, in his introduction to Gemini, emphasised the model’s multimodal nature, designed to generalise and seamlessly combine different types of information. Gemini is most capable and the most flexible model, capable of running efficiently on various platforms, from data centres to mobile devices.

According to rigorous testing, Gemini Ultra, the largest model in the Gemini line-up, has surpassed human experts on the Massive Multitask Language Understanding (MMLU) benchmark, showcasing an unparalleled ability to understand and operate across diverse information types and high capabilities in understanding a wide range of subjects, from math and physics to history and ethics.

Google has invested significantly in the infrastructure behind Gemini, training it on AI-optimised systems using Tensor Processing Units (TPUs) v4 and v5e. The result is a model that runs significantly faster than its predecessors, providing reliability, scalability, and efficiency.

Gemini 1.0 showcases advanced coding capabilities, understanding and generating high-quality code in popular programming languages such as Python, Java, C++, and Go. The model’s prowess in coding extends to competitive programming, as demonstrated by the development of AlphaCode 2, a more advanced code generation system.

To further accelerate Gemini’s development, Google announced the Cloud TPU v5p, the most powerful TPU system to date, designed for training cutting-edge AI models. This technology will empower developers and enterprise customers to train large-scale generative AI models faster and more efficiently.

Highlighting Google’s commitment to responsible AI development, Gemini undergoes the most comprehensive safety evaluations. Safeguards have been implemented to address potential risks associated with the model’s multimodal capabilities, including bias, toxicity, cyber-offence, persuasion, and autonomy.

Gemini 1.0 is set to roll out across a range of Google products and platforms. Gemini Pro will be integrated into products like Bard, providing advanced reasoning, planning, and understanding features. The model will also power new features in Pixel 8 Pro, bringing Gemini Nano to on-device tasks.

Developers and enterprise customers can access Gemini Pro through the Gemini API in Google AI Studio or Google Cloud Vertex AI starting from December 13. Gemini Ultra, the largest model, will undergo additional safety checks before becoming broadly available early next year.