Early this morning.Google The I/O 2025 conference was officially held. At the conference, Google announced several heavy hitters such as the Gemini 2.5 Pro/Flash model, Imagen 4, Android XR glasses, and more.
Multimodal AI assistant 'Project Astra':
Google has brought its multimodal AI assistant, Project Astra.
Google wants to make Gemini a "model of the world" that can both plan, understand, and simulate aspects of the real world, a direction that Google DeepMind CEO Demis Hassabis says is one of the core concepts of Project Astra.
Google is also exploring ways to leverage Agent capabilities to help people multitask, and Project Mariner is one of them, with the ability to do up to ten tasks at once, such as looking up information, making reservations, shopping, and researching.
Android XR products are officially released:
It was the first Android platform built in the Gemini era.
Project Aura, Google's first lightweight smart glasses product with consumer AR glasses brand Xreal, is based on the OST (Optical See through) technology route.
Project Aura will be powered by a Qualcomm Snapdragon XR chip; with two cameras, left and right, and a center-mounted module of other visual sensors for taking photos, videos, and visual/agent intelligence body capabilities that support Gemini (Google Project Astra).
End-to-end AI search experience - AI Mode:
Google's most powerful AI search form ever not only features more advanced reasoning and multimodal understanding, but also supports deep exploration through contextual follow up questions and web links.
Google is bringing Project Astra's multimodal capabilities to search, allowing users to simply turn on their camera and ask questions and get feedback in real time through the screen.
AI Mode also has powerful personalization capabilities, giving customized suggestions based on user contextual preferences, as well as generating charts and visualizations.
Gemini 2.5 family update:
The new version of Gemini 2.5 Pro introduces the "Deep Think" reasoning enhancement mode, which allows for a deeper understanding of the problem context by considering multiple hypotheses before generating an answer. This feature has been ranked as a leader in the 2025 U.S. Math Olympiad (USAMO), LiveCodeBench (a programming benchmark), and scored 84.0% in MMMU (Testing Multimodal Reasoning), outperforming older versions of Gemini 2.5 Pro and OpenAI o3.
The new version of Gemini 2.5 Flash has improvements in key benchmarks such as inference, multimodality, code and long contexts, as well as being more efficient, with a reduction in tokens used in the evaluation from 20% to 30%.
A new subscription service, Google AI Ultra, up to $1,800:
The program provides professional users with unlimited access to Google's most powerful models and advanced features for professionals such as filmmakers, developers, creative workers, etc. for a monthly fee of $249.99 (Rs. 1,804).