Google AI News & Updates

Google Deploys Veo 3 Video Generation AI Model to Global Gemini Users

Google has rolled out its Veo 3 video generation model to Gemini users in over 159 countries, allowing paid subscribers to create 8-second videos from text prompts. The service is limited to 3 videos per day for AI Pro plan subscribers, with image-to-video capabilities planned for future release.

Google Launches Open-Source Gemini CLI Tool for Developer Terminals

Google has launched Gemini CLI, an open-source agentic AI tool that runs locally in developer terminals and connects Gemini AI models to local codebases. The tool allows developers to make natural language requests for code explanation, feature writing, debugging, and other tasks beyond coding. Google is offering generous usage limits and open-sourcing the tool under Apache 2.0 license to encourage adoption and compete with similar tools from OpenAI and Anthropic.

Google Launches Real-Time Voice Conversations with AI-Powered Search

Google has introduced Search Live, enabling back-and-forth voice conversations with its AI Mode search feature using a custom version of Gemini. Users can now engage in free-flowing voice dialogues with Google Search, receiving AI-generated audio responses and exploring web links conversationally. The feature supports multitasking and background operation, with plans to add real-time camera-based queries in the future.

Major AI Companies Withdraw from Scale AI Partnership Following Meta's Large Investment

Google is reportedly planning to end its $200 million contract with Scale AI, with Microsoft and OpenAI also pulling back from the data annotation startup. This withdrawal follows Meta's $14.3 billion investment for a 49% stake in Scale AI, with Scale's CEO joining Meta to develop "superintelligence."

Google Launches AI Edge Gallery App for Local Model Execution on Mobile Devices

Google has quietly released an experimental app called AI Edge Gallery that allows users to download and run AI models from Hugging Face directly on their Android phones without internet connectivity. The app enables local execution of various AI tasks including image generation, question answering, and code editing using models like Google's Gemma 3n. The app is currently in alpha and will soon be available for iOS, with performance varying based on device hardware and model size.

Google Expands Project Mariner AI Agent to Handle Multiple Web-Browsing Tasks Simultaneously

Google is rolling out Project Mariner, an experimental AI agent that browses websites and completes tasks like purchasing tickets or groceries without users visiting sites directly. The updated version runs on cloud virtual machines and can handle up to 10 tasks simultaneously, addressing previous limitations that required users to remain idle while the agent worked.

Google I/O 2025 to Showcase AI Advancements Across Product Lines

Google's upcoming developer conference, Google I/O 2025, will be held on May 20-21 with a strong focus on artificial intelligence. The event will feature presentations from CEO Sundar Pichai and DeepMind CEO Demis Hassabis, highlighting updates to Google's Gemini AI models, Project Astra, and AI integration across Google's product ecosystem including Search, Cloud, Android, and Waymo.

Microsoft Adopts Google's Agent2Agent Protocol for AI Communication

Microsoft has announced support for Google's Agent2Agent (A2A) protocol in its Azure AI Foundry and Copilot Studio platforms. The A2A protocol enables AI agents from different providers to communicate and collaborate across clouds, apps, and services, allowing developers to build complex multi-agent workflows while maintaining governance standards.

Google Releases Enhanced Gemini 2.5 Pro Model with Improved Coding Capabilities

Google has launched Gemini 2.5 Pro Preview (I/O edition), an updated AI model with significantly improved coding and web app development capabilities. The model tops several benchmarks including the WebDev Arena Leaderboard and achieves 84.8% on the VideoMME benchmark for video understanding.

Google's Gemini 2.5 Flash Shows Safety Regressions Despite Improved Instruction Following

Google has disclosed in a technical report that its recent Gemini 2.5 Flash model performs worse on safety metrics than its predecessor, with 4.1% regression in text-to-text safety and 9.6% in image-to-text safety. The company attributes this partly to the model's improved instruction-following capabilities, even when those instructions involve sensitive content, reflecting an industry-wide trend of making AI models more permissive in responding to controversial topics.