Google announces smart glasses and bets on AI that talks and does ‘anything’

The future of artificial intelligence for Google lies in applications that talk in real time and are increasingly autonomous, show the big tech launches this Tuesday (19).

Demis Hassabis, who chairs the company’s advanced artificial intelligence arm, said the company’s new AI model, Gemini Omni, is capable of “doing anything” by receiving “commands any way,” with a focus on video generation.

In practice, the technology will receive commands through text, image, speech and video for almost instantaneous interactions, such as making an edit, adding characters or objects. Examples released by the company show ultra-realistic scenes, with consistent reflections and physics, generated by AI

“That’s always been our goal with Gemini,” said Hassabis during Google I/O, the company’s main event for programmers.

There is still no release date for Gemini Omni, which will be available, in a limited version, to Google AI Pro and Ultra subscribers through the Gemini, Google Flow, YouTube Shorts and YouTube Create applications.

Hassabis also compared the new product with the Nano Banana, Genie and Veo tools. According to him, the models are capable of generating realistic videos and simulations, with notions of reality and physics, but still with some limitations.

The Omni goes a step further: it can represent more complex ideas, like gravity and kinetic energy – which the previous three models couldn’t understand.

A future in which people talk to AI also appears in the launch of Google’s first smart glasses, 11 years after the commercial failure of Google Glass (ended in 2015).

The new product, aimed at the niche dominated by Meta’s Ray-Bans, is born from a partnership with Samsung and the glasses brands Warby Parker and Gentle Monster, bringing together high quality lenses, camera and speakers.

The smart glasses will be available in two designs and will allow the user to “go hands-free and head up,” said Shahram Izadi, a Google executive who led the project. The launch is scheduled for the second half of this year.

The $4.7 trillion technology giant hopes to make up ground against rival AI startups by leveraging its ubiquitous search engine, suite of products and vast database to attract billions of users.

In what he called the biggest transformation in the last 25 years, he added more options for AI searches to Search, aimed at more complex questions, and even included an easier programming tool.

The company also announced a new assistant called Spark, which, according to the company, can help organize users’ personal lives, reduce bureaucracy and automate repetitive tasks.

Users will be able to choose to allow Spark to remember past actions and access data from other products, such as Gmail and Maps, to personalize services.

Spark will be available in the Google search bar. Users will be able to delegate the task of scouring the internet unsupervised to the agent to make purchases, bring news updates, find specific products and book events.

Hassabis also positioned Google in relation to big tech’s main competitors. According to him, the performance of the Gemini 3.5 model, which reaches the public this Tuesday, is comparable to Claude Code, from Anthropic, and Codex, from OpenAI.

Google CEO Sundar Pichai said that the platform is faster and is twice as cheap as competing models.

The executive also said that Gemini now has 900 million monthly users – more than double in a year. The AI ​​Summaries feature in Search now has 2.5 billion monthly users, while AI Mode has about 1 billion, it said.

The demand for supercomputers to run the latest AI models means that the market’s main suppliers need, in some cases, to ration customer access. Suppliers of servers for data centers used in these operations say they are unable to deliver all the orders that have arrived since 2024.

Google hopes that its latest proprietary chips will offer greater capacity to operate its models at a lower computational cost.
A Pro version of Gemini 3.5 will be released next month, consisting of a more robust and capable language model than the Flash version.

Hassabis said artificial general intelligence (AGI), AI technology that can outperform humans at most tasks, is “on the horizon.”

However, “it is important that we are clear-eyed about potential challenges and use all the tools at our disposal to ensure the security of our agent systems,” said the Nobel Prize winner.