Google I/O 2025: AI Announcements Roundup

Here's a breakdown of the most exciting AI updates from Google I/O 2025:

Generative Media:

• Veo 3: Google's advanced video generation model is here! Creates videos with sound effects and dialogue. Available for Google AI Ultra subscribers in the US (Gemini app & Flow). Private preview on Vertex AI. Broader rollout coming soon.
• Veo 2 Updates: Reference-powered video for consistent style, camera controls, outpainting, and object manipulation. Some features in Flow now, full set on Vertex AI soon.
• Imagen 4: Richer, more detailed images with improved text rendering. Free in Gemini app, Whisk, Workspace (Slides, Docs, Vids), and Vertex AI. Faster version launching soon.
• Flow: New AI filmmaking tool using Veo, Imagen, and Gemini. Create cinematic clips via natural language. Available to Google AI Pro/Ultra subscribers in the US.
• Lyria 2 & Lyria RealTime: High-fidelity music generation (Vertex AI) and experimental interactive music model (Gemini API/AI Studio) for real-time generative music.

Gemini App:

• Canvas "Create" Button: Turns chats into interactive content (infographics, quizzes, podcasts) in 45 languages.
• Deep Research: Upload files/images. Google Drive/Gmail integration coming soon.
• Gemini Live: Camera/screen sharing now free on Android/iOS (rolling out). Integrates with Calendar, Keep, Maps, Tasks soon.

Subscriptions:

• Google AI Pro ($19.99/month): Available in US and other countries. New features (Flow, Gemini in Chrome) are US-first.
• Google AI Ultra ($249.99/month): Highest usage limits, early access to Veo 3 & Gemini 2.5 Pro Deep Think, Flow, Agent Mode, YouTube Premium, 30TB storage. Available in US. 50% off for 3 months for new users.
• Student Discount: Free school year of Google AI Pro for college students in US, UK, Brazil, Indonesia, and Japan.

Gemini in Chrome & Agent Mode:

• Gemini in Chrome: Summarize, clarify, and get help with webpages (US, English, Google AI Pro/Ultra). Privacy controls in place.
• Agent Mode: (Coming soon, Ultra desktop users). Gemini handles complex online tasks (filtering listings, filling forms, scheduling). Uses MCP protocol & automated navigation.

AI in Search:

• AI Mode: New tab in Google Search (US). Powered by Gemini 2.5. Advanced reasoning, longer queries, multimodal search, instant answers, "Deep Search" (hundreds of searches).
• Future Integrations: Live capabilities from Project Astra, agentic tools from Project Mariner, personal context from Gmail (user controlled).

Gemini 2.5 Models:

• Gemini 2.5 Pro & 2.5 Flash: Leading coding/reasoning benchmarks. 2.5 Flash has a new preview with improvements. General availability in June 2025.
• Gemini 2.5 Pro Deep Think: Experimental enhanced reasoning mode. Parallel thinking techniques for complex tasks. Launching to trusted testers (Gemini API), then general rollout. Users control "thinking budget."
• Model Context Protocol (MCP): Natively supported in Gemini API/SDK for easier agent/tool integration.
• Thought Summaries: Step-by-step explanations of Gemini's reasoning and tool use (Gemini API & Vertex AI).

Rebranded Projects:

• Project Starline -> Google Beam: AI-powered 3D video calling. Launching with HP and other enterprise partners.
• Project Astra -> Gemini Live: Real-time camera/screen sharing. Free on Android/iOS.
• Project Mariner -> Agent Mode: Agentic computer use (multitasking, browser automation). Available to Ultra subscribers in US, coming to developers (Gemini API/Vertex AI).

Open Models and Developer Tools:

• Gemma 3n: Efficient multimodal open model for fast, low-memory devices. Supports text, audio, image, multilingual input. Preview for developers on AI Studio/AI Edge.
• Jules: Asynchronous coding agent (Gemini 2.5 Pro). Public beta (free for now). Handles coding tasks within GitHub. Concurrent tasks, audio changelog.
• Gemini Diffusion: Experimental research model for fast text generation (5x faster than previous models). Developer preview via waitlist.
• SynthID Detector: Portal for checking if content was generated using Google's AI tools. Rolling out to early testers via waitlist