Gemini 1.5 is notable for being a true mixture of experts, activating specific paths for different data types, making it more energy efficient compared to models that use modular approaches.

The Gemini model is being leveraged to enhance AI capabilities in projects like Mariner, focusing on long context and memory.

The increase in tokens processed by Gemini reflects the limitless human curiosity, with a significant rise from 9.7 trillion to 480 trillion tokens per month in just a year.

The Gemini Developer API introduced many new features, including thought summaries directly in the API, native audio output for better voice control, and a URL context tool that allows models to analyze web links, enhancing the building blocks for new capabilities.

The investment in the Gemini program has focused on integrating various capabilities quickly, allowing the model to handle tasks like coding, text generation, and image generation all in one place.

Gemini was built to be natively multimodal, meaning it can take in images, audio, video, and more, which is essential for progressing towards Artificial General Intelligence (AGI).

Gemini 2.5 is our most secure model yet, with strengthened protections against security threats and improved transparency through thought summaries.

The Gemini 2.0 Pro model has impressed me with its capabilities, particularly its ability to process large amounts of information efficiently.

Over 7 million developers have built with the Gemini API, reflecting over 5x growth since last I/O, and Gemini usage on Vertex AI has increased more than 40 times since last year.