At Google I/O, Google unveiled a slew of updates and new AI-focused products. Objective: to win back developers and establish itself among individuals.
2026 is indeed the year of personal intelligence. This is what emerges from Google’s announcements at its annual I/O conference in Mountain View, this Tuesday, May 19. AI now infuses the entire Google software stack, with a clear objective: to personalize the user experience as much as possible. Back to the main announcements.
Gemini Omni, a new model family
This is the flagship announcement of the conference. Expected for several days and previewed on the web, Gemini Omni is a new family of natively multimodal AI models. Sound, image, video, text… AI is capable of processing all modalities jointly to produce a coherent result. Omni’s signature generation capability? The video. With Gemini Omni, Google promises that it becomes possible to edit your own videos with AI, opening up a truly new experience.
For the moment, Gemini Omni only allows you to generate output videos. The long-term objective, however, is to support any type of output modality.
Gemini 3.5 Flash, more powerful, faster
This is the announcement that will certainly delight developers the most. On the occasion of I/O, Google announces the arrival of Gemini 3.5, its new family of state-of-the-art LLMs. For now, Google is only releasing Flash version 3.5. In the benchmarks it would be superior to 3.1 Pro in code, agent orchestration and tools calings. The real strength of the model? It is up to 4 times faster (250 tokens / second) than other competing frontier models (Claude Opus 4.7 or GPT-5.4) for half the price. Quite strong, very Google.
The model is available everywhere in the Google ecosystem, from Google AI Studio to the Gemini app, Antigravity or even in its AI Mode. Gemini 3.5 Pro has not yet been announced, but Google promises to release the model next month. We must expect, here again, a new intelligence gap.
Gemini Spark, Google’s answer to Claude Desktop
The goal of Gemini Spark? That each user has a personal AI agent, like Claude Desktop. Docs, emails, chat… The agent can execute complex tasks in the background. Based on Gemini 3.5, it is capable of creating files completely independently (slides, Excel spreadsheets) without the user having to intervene. The tool supports MCPs and skills. Spark can be searched on PC but also on smartphone and, thanks to an advanced voice mode, can be easily used anywhere.
The agent will be available to Google AI Pro subscribers in the United States starting next week. To function, Spark relies on a virtual machine in the Google cloud, making it unnecessary to leave your machine on continuously.
A new version of Antigravity focused on agentics
With Antigravity 2.0, Google is transforming its IDE into a veritable factory of autonomous agents. The development environment becomes a native desktop application that serves as a central hub for orchestrating AI agents on the same project. The goal: to develop complex projects more quickly and with great precision. Each agent has access to the shell and tools, based on the principle of Claude Code, to carry out its tasks.
To illustrate the capabilities of the platform, Google indicates that Antigravity 2.0 managed to develop the core of a functional operating system in just 12 hours, by mobilizing 93 AI sub-agents and generating 2.6 billion tokens. Will these new features be enough to attract a community of developers that Anthropic has been capturing for several months? Nothing is less certain.
AI glasses for this year
Google is taking advantage of its conference to confirm the arrival in the fall of its first AI glasses under Android XR, a platform co-developed with Samsung and Qualcomm. Two ranges are planned: an audio model incorporating a Gemini voice assistant, and a model equipped with an integrated display, expected later. Google has teamed up with Gentle Monster and Warby Parker for the design part, with frames designed for everyday wear.
In terms of functionality, the audio glasses rely on Gemini to identify what the user is looking at, provide navigation, manage calls and SMS, capture photos and videos that can be edited via Nano Banana, translate spoken and written words in real time, and trigger actions in several steps. The glasses will be Android and iOS compatible. The audio version will be available in the fall.
Other AI news from I/O
From the more than two hours of announcements, here is the other news from I/O 2026:
- Google introduces new mid-tier AI Ultra plan at $100 per month; the most expensive subscription increases to $200 per month, a price now aligned with that of Anthropic.
- AI Mode search is becoming more intelligent and now offers mini-applications generated on the fly according to queries, to better inform users.
- With Expression Neural, Gemini Live – which allows you to converse in real time with AI, becomes more natural and fluid in your exchanges.
- The Gemini Mac app integrates Gemini Spark to automate tasks directly from your computer.
- Google presents Pics, a new Workspace application entirely dedicated to the creation of images, powered by the Nano Banana 2 model.
Google’s strategy? Raid developers and the general public
The density of announcements at this Google I/O says everything about the company’s strategy. On the developer front, the offensive is massive: Google wants to recover the technical teams that left for OpenAI and Anthropic. Antigravity 2.0 contributes to this, but it is above all Gemini 3.5 Flash which constitutes the strong argument, cheaper, more powerful, up to four times faster than the competition. A value proposition that’s hard to ignore.
On the consumer front, the ambition is just as clear: to become the default AI reflex, from research to connected glasses to Gemini Spark. Google is starting from afar: the failures of 2023 have left their mark but its structural assets remain considerable, distribution, infrastructure, data, native integration in Android and Chrome. The risk, on the other hand, is real. By wanting to cover all fronts simultaneously, Google exposes itself to a perception of dispersion. Dozens of products announced in a few hours, model names that proliferate, price offers that overlap: readability takes a hit. The next step will therefore be as important as the announcements themselves: consolidate, merge, simplify. Transforming the essay will require considerable editorial and product effort. There is no shortage of announcements. The execution remains to be demonstrated.
