At the first major developer conference keynote of the summer, Google and Alphabet Inc. CEO Sundar Pichai outlined the company’s expanding artificial intelligence ambitions, highlighting rapid growth in AI usage across Google’s ecosystem.
Pichai revealed that Google is now processing 3.2 quadrillion AI tokens every month, marking a dramatic jump from the 480 trillion tokens reported during last year’s Google I/O keynote.
The increase reflects Google’s aggressive push to integrate AI across consumer products and enterprise platforms over the past year. Recent launches have included deeper AI integration through Gemini Intelligence on Android devices and conversational search features inside Google Maps.
Pichai said Google’s long-standing AI-first strategy continues to shape the company’s direction, pointing to investments across custom chips, research, models, and products used by billions worldwide.
Google introduces Gemini Omni and Gemini 3.5 Flash
This year’s keynote introduced two major AI models, Gemini Omni and Gemini 3.5 Flash, alongside broader efforts to strengthen conversational AI and agent-based workflows.
Koray Kavukcuoglu, chief AI architect at Google, described Gemini Omni as a model that combines reasoning capabilities with creative generation.
The system allows users to input text, images, audio, and video together in order to generate high-quality video content built around Gemini’s understanding of real-world knowledge.
Google plans to integrate Gemini Omni into the Gemini app, Google Flow and YouTube Shorts.
Kavukcuoglu said the model enables more natural video editing through conversational instructions while maintaining visual consistency, realistic physics and continuity between scenes.
Gemini 3.5 Flash focuses on AI agents and coding
Google also placed heavy emphasis on agentic AI systems capable of performing tasks independently.
Jeff Dean, chief scientist at Google DeepMind and Google Research, said Gemini 3.5 Flash combines advanced intelligence with action-driven workflows.
According to Dean, the model is currently Google’s strongest system for agentic tasks and coding-related performance.
Google said Gemini 3.5 Flash scored 76.2 percent on the Terminal bench 2.1 coding benchmark, outperforming earlier Gemini models and Anthropic’s Claude Opus 4.7, though remaining slightly behind OpenAI’s GPT 5.5.
The company also claimed the model leads in areas such as multistep workflows, tool usage, financial analysis, chart interpretation and multimodal reasoning.
Gemini app gets new Spark AI agent
Google additionally announced major upgrades for the Gemini app through a new AI agent called Spark.
The company said the Gemini app now has 900 million users worldwide, up sharply from 400 million users a year ago.
Josh Woodward, vice president of Google Labs and Gemini, described Spark as a shift from a traditional assistant into a more active AI partner capable of completing real tasks under user direction.
Spark is web based and designed to integrate with other Google applications, allowing recurring tasks and background actions even when a user’s device is turned off.
Google said Spark will always operate under user control and will request permission before carrying out sensitive actions such as sending emails or making purchases.
The feature will initially launch in beta for subscribers of Google’s AI Ultra plan in the United States, which currently costs 100 dollars per month.
Google expands AI ecosystem partnerships
Google also announced new model context protocol integrations with several major creative and technology platforms.
The first confirmed partnership involves Canva, with additional integrations planned for companies including Adobe, Xiaomi, Samsung, CapCut, OnePlus and Dropbox.
