What is Google Gemini Ultra?

Google Gemini Ultra is part of Google's new suite of large language models under the Gemini umbrella. It includes three models: Ultra, Pro, and Nano. Gemini Ultra specifically is the model that has achieved noteworthy performance, surpassing benchmarks set by OpenAI's GPT-4 on certain tasks. For example, Gemini Ultra achieved a reported 90.4% on the MMLU benchmark, a common and popular test for language models 1.

Google has positioned Gemini Ultra as an advanced AI capable of various tasks including text, code, audio, image, and video understanding and processing. The models are designed to be natively multimodal, meaning they are pretrained on different types of information and can seamlessly integrate and reason across these various modalities 2.

Gemini Ultra was reportedly set for release after further safety testing, and there was speculation that it would be associated with a subscription product called Bard Advanced, which could serve as a counterpart to premium AI offerings from other providers like OpenAI 3. As of the information available, Gemini Ultra appears to represent a significant step forward in Google's AI capabilities.