Google I/O 2025: 10 Most Exciting Announcements

Google I/O 2025, held on May 20–21, 2025, at the Shoreline Amphitheatre in Mountain View, marked a significant leap in artificial intelligence, developer tools, and cross-platform integration. The event showcased Google’s vision of a “universal AI,” embedding advanced capabilities across its ecosystem.

Google I/O 2025: 10 Biggest Announcements

Here are the ten biggest announcements of Google I/O 2025.

Gemini 2.5 Pro: Deep Think Mode

The Gemini 2.5 Pro model’s introduction of Deep Think Mode at Google I/O 2025 represents a significant step forward in the evolution of AI reasoning capabilities. Traditional language models typically generate responses in a relatively linear fashion—predicting the next token based on prior context without deeply reevaluating alternative solutions. 

In contrast, Deep Think Mode allows Gemini 2.5 Pro to simulate a more deliberative thinking process by internally exploring and comparing multiple possible answers or reasoning paths before producing a final output. The result is a more accurate, contextually aware, and robust response generation process.

By leveraging Deep Think Mode, Gemini can better tackle benchmarks such as MMLU (Massive Multitask Language Understanding), GSM8K (grade school math problems), and HumanEval (code generation and reasoning), where a model’s ability to assess and refine potential answers directly impacts its performance. 

Furthermore, Deep Think Mode aligns with broader trends in AI development focused on meta-cognition—the ability of a model to not just generate answers, but to reflect on and improve them internally before finalizing output. This makes Gemini 2.5 Pro more effective for enterprise applications, complex technical writing, coding tasks, and domains requiring high factual accuracy and logical consistency.

Gemini Ultra: The Pinnacle of AI Access

Google introduced Gemini Ultra at Google I/O 2025. It is an exclusive $249.99/month. The subscription is available in the U.S., offering premium access to Google’s AI-powered services.

Subscribers receive:

• Gemini 2.5 Pro with Deep Think Mode: An advanced reasoning model that evaluates multiple solutions before responding, enhancing decision-making processes.
• Veo 3: A video generation tool capable of creating sound effects, background noises, and dialogue to accompany videos.
• Flow: An AI-powered video editing app designed for filmmakers.
• Enhanced NotebookLM and Whisk: Platforms for note-taking and image remixing with higher usage limits.
• 30TB Storage: Across Google Drive, Photos, and Gmail.
• Agentic Tools: Powered by Project Mariner, enabling autonomous web interactions.
• YouTube Premium: Included with the subscription.

Gemini App Updates

• Gemini Live: Introduced camera and screen-sharing capabilities, allowing near-real-time verbal conversations with Gemini while streaming video from smartphones.
• Deep Research: Updated to allow users to upload private PDFs and images for generating comprehensive research reports.
• Stitch: An AI-powered tool that assists in designing web and mobile app front ends by generating necessary UI elements and code from prompts.
• Jules: Expanded access to this AI agent aimed at helping developers fix bugs in code, understand complex code, create pull requests on GitHub, and handle certain programming tasks.

Veo 3: AI-Driven Video Generation

Veo 3, an upgraded video generation model, can create videos with sound effects, background noises, and dialogue. It offers improved quality over its predecessor, Veo 2, and is accessible to Gemini Ultra subscribers. Veo 3 also supports a wider range of cinematic styles, allowing creators to fine-tune visual tone and pacing. It features enhanced motion consistency, producing smoother and more realistic character movements. Integration with Google’s creative tools makes it easier to incorporate Veo 3 videos into larger multimedia projects.

Imagen 4: Advanced Image Generation

Imagen 4 is a fast and efficient image generation model capable of rendering fine details such as fabrics, water droplets, and animal fur with remarkable precision. It supports both photo-realistic and abstract styles, giving users the flexibility to generate visuals that suit a wide range of creative needs. 

The model accommodates various aspect ratios and produces high-quality images at resolutions of up to 2K. Designed for both speed and visual fidelity, Imagen 4 is well-suited for applications in design, advertising, digital art, and content creation, offering a powerful tool for generating compelling visuals quickly and accurately.

Project Mariner: Autonomous Web Interaction

Project Mariner, developed by Google DeepMind, was also announced at Google I/O 2025. It is an experimental AI agent designed to autonomously browse and interact with websites using Gemini 2.0’s advanced multimodal capabilities. As a Chrome extension, it can interpret and act on text, images, code, and forms, allowing users to complete tasks such as purchasing tickets, filling out forms, or shopping online through simple conversations with Gemini—without needing to visit third-party websites themselves. It automates web navigation and task execution while ensuring user control by requesting confirmation for sensitive actions like purchases.

Announced at Google I/O 2025, Project Mariner is part of a broader push toward AI agents that perform complex, real-world tasks. It includes features like multitasking (handling up to 10 tasks simultaneously), a “Teach and Repeat” function that learns tasks from user demonstrations, and strong performance on the WebVoyager benchmark. Currently in limited testing, it is expected to see broader availability by summer 2025. Project Mariner is a key step in DeepMind’s vision for AI-enhanced productivity through seamless browser-based automation.

Project Astra: Real-Time Multimodal AI

Project Astra is Google’s low-latency, multimodal AI experience that powers new features in Search, the Gemini AI app, and products from third-party developers was also showcased at Google I/O 2025. It enables near-real-time, multimodal AI capabilities and is being developed in collaboration with partners like Samsung and Warby Parker.

AI Mode in Google Search

AI Mode is an experimental feature in Google Search that allows users to ask complex, multi-part questions via an AI interface. It was officially announced at Google I/O 2025. It supports complex data in sports and finance queries and offers “try it on” options for apparel. Later this summer, Search Live will enable users to ask questions based on what their phone’s camera is seeing in real-time

Beam 3D Teleconferencing

Beam, previously known as Project Starline, was also revealed at Google I/O 2025. It uses a combination of software and hardware, including a six-camera array and custom light field display, to let users converse with someone as if they were in the same meeting room. It boasts near-perfect millimeter-level head tracking and 60fps video streaming. When used with Google Meet, Beam provides an AI-powered real-time speech translation feature that preserves the original speaker’s voice, tone, and expressions 

Wear OS 6

Wear OS 6 got a unified font for tiles for a cleaner app look at Google I/O 2025. The Pixel Watches are getting dynamic theming that syncs app colors with watch faces. The core promise of the new design reference platform is to let developers build better customization in apps along with seamless transitions. Google is releasing a design guideline for developers along with Figma design files 

Which of these Google I/O 2025 announcements excited you the most and why? Share it with us in the comments section below.