Sundar Pichai Unveils Google’s Full-Scale Gemini AI Push At I/O 2025

Google CEO Sundar Pichai opened the 2025 I/O developer conference by declaring a “new phase of the platform shift,” as the company unveiled sweeping advances across its AI ecosystem.

He said Gemini now processes 480 trillion tokens per month, a 50-fold increase from 9.7 trillion just a year ago. 

The model powers over 400 million monthly active users, and Google Search’s AI Overviews now reach 1.5 billion users each month— a scale Pichai said shows Google is bringing generative AI “to more people than anyone.”

He also announced that Google is commercializing Project Starline, its once-experimental 3D video conferencing platform, under a new name – Google Beam. 

The system uses six cameras and AI to generate a 3D-like display with real-time 60 fps rendering. HP will ship the first Beam devices later this year. 

Google Meet is also adding real-time English-to-Spanish translation today, with more languages on the way.

Gemini Live, which enables more natural, back-and-forth voice interactions, is rolling out to Android and iOS starting on Tuesday. It will be available to subscribers first and enterprises later this year. 

Personalized responses on Gmail are also set to get an upgrade with Gemini’s “personal context” feature, which allows Gemini to write smart replies that “sound like you.” 

“With your permission,” Gemini can scan information from a user’s Gmail inbox and Google Drive to better match their tone and style. 

The improvements build on Google’s “contextual” upgrade to smart replies introduced last year. Pichai said the feature is likely to be introduced this summer. 

The company also introduced Gemini 2.5 Flash, its most efficient model to date, optimized for speed and cost. Flash will be available in early June, with the more powerful 2.5 Pro version to follow. 

Agent-based computing also took center stage. Google is integrating the technology behind Project Mariner, its task-executing AI prototype introduced last year, into its consumer products. 

Mariner can manage up to ten simultaneous tasks, learn from user instructions, and repeat complex workflows. 

This functionality, now branded as Agent Mode, will soon appear in Gemini, Chrome, and Search. Developer access will be available through the Gemini API this summer.

Jules, Google’s AI-powered coding agent, which was unveiled in December, is entering public beta.

Pichai also announced that AI Mode in Google Search is rolling out to everyone in the U.S. starting Tuesday. AI Mode will now power AI overviews, which Pichai called “one of the most successful launches for Google Search in the past decade,” and has led to 10% growth in queries that show overviews.

Google-parent Alphabet’s (GOOG/GOOGL) shares were down 1.1% in afternoon trade. Alphabet's stock is down more than 13% this year, and has fallen over 6% in the last 12 months.

For updates and corrections, email newsroom[at]stocktwits[dot]com.

Read also: Apple iPhone And Other Smartphone Imports From China Hit 14-Year Low Amid Tariff Tensions