At the recent Google I/O conference, Google demonstrated its massive AI scale, with CEO Demis Hassabis revealing that the Gemini App has surpassed 900 million monthly active users, processing 32 quadrillion tokens per month. Gemini has transitioned from a standalone app to the foundational AI engine for the entire Google ecosystem.
The spotlight first fell on Gemini Omni, described by DeepMind's CEO as a model capable of generating anything from any input. Positioned as a "world model," it integrates reasoning with generative media to understand physical properties like kinetics and gravity. Demonstrations showed the model generating claymation videos for complex biological concepts and enabling natural video editing through dialogue, such as transforming objects or altering cinematic lighting in real-time.
Google also introduced Gemini 3.5 Flash, a model optimized for speed, cost-efficiency, and agentic workflows. It outperforms previous models in coding benchmarks and complex tasks, boasting output speeds 4 to 12 times faster than its predecessors. Google noted that internal development now processes over 3 trillion tokens daily, creating a feedback loop for continuous model improvement.
Complementing the models is Antigravity 2.0, now a standalone desktop application focusing on an "agent-first" philosophy. In a high-intensity demo, 93 parallel agents spent 12 hours and 2.6 billion tokens to build a functional operating system from scratch. The system was even capable of running Doom after agents autonomously generated and fixed video drivers, compressing weeks of engineering into hours.
Finally, Google Search is being reimagined as an Information Agent. Upgraded to Gemini 3.5, AI Mode now supports multimodal inputs and preserves context across queries. A new "Information Agent" feature coming this summer will allow users to track specific data like stock metrics or real estate listings autonomously. Search is evolving from static results into dynamic, interactive interfaces generated in response to complex user needs.
