Digital Economy

From Google Ultra to the landing of Gemini on Chrome. What's new at Google I/O

Today's announcements at Google I/O (its well-known developer conference) are all about AI

20 May 2025

4' min read

Personalised, universal, ubiquitous. Today, Google is relaunching a mission shared by many artificial intelligence bigwigs, OpenAI in primis: they want to make it our personal and universal assistant at the same time, for everyday life, work. With more intelligence, more functions, more autonomy, more understanding of our needs. Today's announcements at the Google I/O (its well-known conference dedicated to developers) are all about AI, and in part trace what other biggies are also doing (see the increasingly sophisticated reasoning models). Google puts its own spin on agents, multimodality and video, and the promise of integration with the search engine and its other services.

Some of the more advanced things - such as the video generator, now also with sound, Veo 3 and the landing of Gemini on Chrome - are reserved for those who subscribe to the new AI Ultra plan at $249.99 per month (similar to Chatgpt Pro). In any case, for now only in the US.

Gemini 2.5

Many Gemini 2.5 improvements announced. Flash is now the default mode.

Native audio output for more natural conversation and advanced security measures are coming. Gemini 2.5 Pro will be enhanced - Google announces - with Deep Think, an experimental advanced reasoning mode for highly complex mathematical and programming operations.

The 2.5 Flash is the most reliable and efficient model, designed for speed and low cost, which has now been refined in almost every aspect. It has improved in key benchmarks for reasoning, multimodality, programming and long context while becoming even more efficient, as it uses 20-30% fewer tokens.

The 2.5 Pro, on the other hand, is listed by Google as a leader in certain rankings, such as the WebDev Arena and LMArena programming rankings. In addition, 2.5 Pro is now also Google's leading learning model. "In direct comparisons evaluating its pedagogy and effectiveness, educators and experts preferred Gemini 2.5 Pro," Google announced.

Universal assistant AI app, also with agents

"Our ultimate vision is to turn the Gemini app into a universal AI assistant that will take care of everyday tasks for us, handle the most tedious administrative tasks and offer us new and pleasant recommendations - making us more productive and enriching our lives," Google announced.

A goal that will have to pass through numerous research developments, now at a prototype stage, under the name of Project Mariner and Project Astra .

Project Mariner now includes an agent system capable of performing up to ten tasks simultaneously. These agents can help you search for information, make reservations, make purchases, conduct research and more - all at the same time. The updated version of Project Mariner is available to Google AI Ultra subscribers in the United States.

Google is also integrating Project Astra's Live functionality into Search in the US, so that you can interact with Search in real time about what you see using your camera. "Stuck on a project? Just tap the "Live" icon in AI Mode, point the camera, ask your question and get an explanation, with links to other resources you can explore in more detail," Google explains.

At the same time, Google makes AI more personalised. 'With your permission, it will integrate with Google services like Maps, Calendar, Tasks and Keep (and others in the future) for more in-depth daily assistance,' he explains. A trend, this of personalisation based on our data, which also concerns OpenAI, Meta, Microsoft. Google has on its side the possibility of using what we do on those famous services.

New Creative Tools

Google today presents Veo 3 and Imagen 4, its models for generating videos and images. Veo 3 not only improves on the quality of Veo 2, but is also able, for the first time, to create videos with sound: traffic noise in the background on a city street, birds singing in a park, even people talking.

Imagen 4 can create images in a wide range of proportions and up to 2k resolution, for printing or presentations. Spelling and typography have been significantly improved, which makes it easier to create greeting cards, posters and even comics. An evolution that OpenAI has already reached with Gpt4.

Also arriving is Flow, our new AI-based film production tool for creating narrative elements (cast, locations, objects, styles) with natural language, all in one place.

Search with artificial intelligence

AI Overview represents one of the most successful launches in Google Search in the past decade, the company says today. "In our core markets, such as the US and India, AI Overview is contributing to a more than 10 per cent increase in Google usage for the types of queries that trigger this functionality."

So Google is moving forward. Today it extends AI Mode to all US users (no more registration required).

"AI Mode is where we will introduce Gemini's most advanced capabilities for the first time, and it is also a glimpse of what lies ahead. As we receive feedback, many features and capabilities will move from AI Mode directly into the main Search experience. Starting this week, we will introduce a customised version of Gemini 2.5, in both AI Mode and AI Overview search in the US."

"For questions that require an even more comprehensive answer, we are introducing in-depth search capabilities in AI Mode through Deep Search.

In addition, it introduces 'live' functions ('with Search Live you can converse with Search about what you see in real time, using the camera') and Mariner's agentics in Ai Mode search.

Ai Mode will become more personalised, with our context information, and will have a shopping mode. "See how a garment fits and ask our agent checkout function to make the purchase when the price is right".

All this for now only in the US.

The ambition, with tools and prerogatives specific to Google, is that: AI as a total universal tool for everyone. After all, there is no other way to pay back billionaire investments. The other big tech companies are also making them, however, and have similar ambitions. In the end, perhaps only one will prevail - how many universal AIs are there in our lives, after all? Right now, the landscape is confusing. But the big guys, as the AI market matures, will have to find better ways to differentiate themselves from each other.