Keep pulling the thread on Jeff Dean.
The Gemini Ultra model achieved state-of-the-art performance on 30 out of 32 benchmarks evaluated by the Gemini Team.
The Gemini Ultra model is the first model to achieve human-expert performance on the MMLU benchmark.
The Gemini Ultra model improved the state of the art in all 20 of the multimodal benchmarks examined by the Gemini Team.
Gemini is a family of multimodal models with capabilities across image, audio, video, and text understanding.
The Gemini family of models consists of three sizes: Ultra, Pro, and Nano.
The Gemini Nano model is suitable for on-device, memory-constrained use-cases.
Gemini models are deployed to users through services including Gemini, Gemini Advanced, Google AI Studio, and Cloud Vertex AI.