Google I/O 2025 released a series of important updates covering AI, Android, Chrome, and other areas.
Alphabet $Alphabet-A (GOOGL.US)$ The grandest annual developer conference, Google I/O 2025, will be held this Tuesday and Wednesday at the Shoreline Amphitheatre in Mountain View, California. This is the stage for showcasing updates across its entire product line, covering Android, Chrome, Google Search, YouTube, and of course, the essential AI chatbot Gemini, among many other areas.
Google also held a separate event specifically for the Android update. The company announced several new features, including new ways to locate lost Android phones and other items, device-level security features added to the Advanced Protection program, security tools to prevent scams and theft, and the newly designed Material 3 Expressive.
Here are the highlights announced at Google I/O 2025:
Gemini Ultra
Gemini Ultra (currently available only in the U.S.) offers 'the highest level of access' to Google's AI applications and services for a monthly fee of $249.99. The package includes the Veo 3 video generator, the newly launched video editing tool Flow, and the powerful AI feature Gemini 2.5 Pro's Deep Think mode, which has not yet been released.
Subscribers of Gemini Ultra will also receive enhanced quotas for NotebookLM and the image mixing application Whisk, as well as use of the Gemini chatbot in Chrome, multiple 'smart agent' tools based on Project Mariner technology, YouTube Premium, and 30TB of storage in Google Drive, Photos, and Gmail.
Deep Think mode of Gemini 2.5 Pro
Deep Think is an "enhanced reasoning mode" of the Gemini 2.5 Pro model, capable of synthesizing multiple answers before responding, thereby improving the model's performance in certain benchmark tests.
Google has not detailed its specific principles, but it may be similar to OpenAI's o1-pro or the upcoming o3-pro, possessing the ability to search and integrate optimal solutions.
Deep Think is currently accessible to "trusted testers" through the Gemini API. Google stated it will undergo additional security assessments before a wider release.
Veo 3 video generation AI
Google claims that Veo 3 can generate sound effects, background noise, and even voice dubbing. The image quality is also superior to the previous generation Veo 2.
Veo 3 will launch the Gemini chatbot application starting Tuesday, accessible only to Gemini Ultra subscribers, allowing them to generate video content via text or image prompts.
Imagen 4 image generation AI
Imagen 4 is faster than Imagen 3, and a version ten times faster than Imagen 3 will be released in the future. It can generate "fine details" such as fabrics, water droplets, and animal fur, supporting both realistic and abstract styles, with images reaching up to 2K resolution and various aspect ratios.
Veo 3 and Imagen 4 will provide core support for the video creation tool Flow.
Gemini application update.
Google announced that the monthly active users of the Gemini series applications have exceeded 0.4 billion.
Gemini Live will open up camera and screen sharing features to all iOS and Android users this week, allowing users to interact with AI through near real-time voice communication and share their phone screens in real time using Project Astra technology.
In the coming weeks, Gemini Live will also deeply integrate with other Google ecosystem applications, such as calling Google Maps for navigation, creating Calendar events, and managing task lists.
In addition, the Deep Research feature has been upgraded, allowing users to upload private PDFs and images to generate research reports.
Stitch.
Stitch is an AI tool that can be used to design web pages and mobile app front ends, where users can generate UI elements and HTML and CSS code with just a few sentences or an image.
Although Stitch's functionality is somewhat limited compared to some 'visual programming' tools, it offers a high degree of customization.
At the same time, Google has expanded the usage scope of the developer AI assistant Jules, which can help understand complex code, create Pull Requests on GitHub, handle code backlog tasks, and more.
Project Mariner
Project Mariner is Google's experimental AI agent that can access and operate websites on behalf of users. It has now been updated to support handling nearly ten tasks simultaneously and is starting to be opened to some users.
For example, users can complete tasks such as purchasing tickets or online shopping by simply chatting with the AI, without needing to open third-party websites.
Project Astra
Project Astra is a low-latency multimodal AI project launched by Google DeepMind, which will support search, Gemini applications, and third-party products. Alphabet-A is also collaborating with Samsung,$Warby Parker (WRBY.US)$and other companies to develop Project Astra glasses, but the release date has not yet been announced.
AI Mode
Google is launching AI Mode in the United States - an experimental search feature that allows users to ask complex multi-part questions through an AI interface.
AI Mode can handle complex data queries related to sports and finance, and also offers a "try on" feature for clothing. Search Live, launching later this summer, will support visually-based search inquiries using real-time footage from mobile phone cameras.
Gmail is the first application to support personalized context handling.
Beam 3D video conferencing
Beam (formerly known as Starline) combines a six-camera array with a customized light-field display to make remote meetings feel like face-to-face interactions. Its AI model synthesizes video streams from different angles into a 3D rendered image.
Beam achieves millimeter-level head tracking and video stream transmission at 60 frames per second. When used in conjunction with Google Meet, it can also provide real-time AI voice translation while preserving the original speaker's tone, pitch, and facial expressions.
Google Meet itself will also support a real-time voice translation feature.
More AI updates
Gemini will be integrated into the Chrome browser as a new AI browsing assistant to help users quickly understand page content and complete tasks.
Gemma 3n is an AI model optimized for smartphones, laptops, and tablets, with previews starting Tuesday, supporting audio, text, images, and video processing.
Google has also brought numerous AI office updates to Gmail, Docs, and Vids. Gmail will introduce personalized smart replies and inbox cleanup features, while Vids will enhance content creation and editing capabilities.
NotebookLM will add a video overview feature, while Google launched SynthID Detector—a platform for AI content recognition based on SynthID watermark technology. The music generation model Lyria RealTime will also be available for use via API.
Wear OS 6
Wear OS 6 introduces a unified font to enhance interface consistency; the Pixel Watch will support theme color synchronization to enhance dynamic aesthetics.
The new design platform will help developers build richer personalized applications and achieve seamless interface transitions. Google will provide design guidelines and Figma template files for developers.
Google Play
Google has added multiple tools for Android developers in the Play Store, including subscription management, content previews (such as audio snippets), and smoother payment processes.
American users can access the 'Theme Browsing' page to quickly discover apps related to films and television works. Developers will also receive exclusive testing and publishing pages as well as app release monitoring tools. In the event of serious issues, developers can pause app releases.
The subscription tools have also been upgraded to support multi-product checkout. Developers can sell additional services under the main subscription for unified billing.
Android Studio
Android Studio will integrate several new AI features, including the 'Journeys' (AI development agent process) in conjunction with Gemini 2.5 Pro and the 'Agent Mode' automated development function.
Additionally, the 'Crash Insights' feature in the App Quality Insights panel will also be supported by Gemini to help analyze source code, identify causes of crashes, and provide repair suggestions.
Editor/rocky
