At Google I/O 2025, Google announced a slew of new AI models, tools, and subscription services. Here are some of the most notable updates and their release dates.
**Generative media**
- Veo 3, Google's most advanced video generation model capable of creating videos with sound effects and even dialog, is now available to Google AI Ultra subscribers in the U.S. and can be accessed through the Gemini app and Flow, as well as privately previewed on Vertex AI, with a wider rollout in the coming weeks.
- Veo 2 adds new features such as reference-driven video (to maintain consistent styles and characters), camera controls for precise lens adjustments, augmented aspect ratios, and object add/remove, with some of the new features now available in Flow, and the full range of features coming soon in Vertex AI.
- Imagen 4 generates richer, more detailed and accurate images, with improved text rendering and cueing results, and is now available for free in the Gemini app, Whisk, Workspace (Slides, Docs, Vids) and Vertex AI, with a faster version coming soon.
- Flow, a new AI movie-making tool that allows users to create movie clips using Veo, Imagen and Gemini through natural language and asset management, is now available to Google AI Pro and Ultra subscribers in the US.
- Lyria 2, Google's music generation model, is now live at Vertex AI and supports high-fidelity adaptive music generation, while Lyria RealTime is available as an experimental interactive music model offering real-time generation and performance-generated music via the Gemini API and Google AI Studio.
Gemini Applications
- Canvas has a new one-click "Create" button that makes it easy to turn chats into interactive content such as infographics, quizzes, and podcasts in 45 languages, Deep Research now allows for uploading files and images, and Google Drive and Gmail integration is coming soon.
- Gemini Live's camera and screen sharing features are now available for free on Android and iOS (rollout underway) and will soon be integrated with apps like Google Calendar, Keep, Maps and Tasks.
subscription service
- Google AI Pro ($19.99 per month) is already available in the U.S. and other countries, but some of the newest features (such as Flow or Gemini in Chrome) are rolling out in the U.S. first, followed by a wider rollout.
- Google AI Ultra ($249.99 per month, 50% off for the first three months for new subscribers) offers the highest usage limits, earliest access to premium models (such as Veo 3 and Gemini 2.5 Pro Deep Think), the highest limit of Flow features, and exclusive access to Agent Mode, and also includes YouTube Premium and 30TB of storage, is available now in the U.S. with more countries coming soon.
- College students in the U.S., U.K., Brazil, Indonesia and Japan can get one academic year of Google AI Pro services for free.
---
OpenAI has made a major update to its Responses API with the addition of remote MCP server support. This feature allows users to process and manage data more efficiently via the cloud.
In addition, OpenAI has introduced real-time streaming image generation and multi-round editing capabilities at a cost of $5.00 per 1 million text input tokens, $10.00 per 1 million image input tokens, and $40.00 per 1 million image output tokens. This will greatly enhance the efficiency and quality of image content generation for a variety of creative design and content production scenarios.
Code InterpreterThe functionality has also been enhanced to help developers run code and get results more efficiently at a cost of $0.03 per container.
For file search, OpenAI has boosted its performance to cost $0.10 per GB of vector storage per day and $2.50 per 1,000 tool calls, enabling users to find the information they need more quickly and accurately.
These new features are now available in GPT-4o, GPT-4.1, and OpenAI's o-series models. The update also includes features such as background mode, inference summarization, and encrypted inference terms, further enhancing the security and utility of the models.
---
OpenAIDevs has announced the introduction of a number of new features in the Responses API, including support for remote MCP servers, image generation, code interpreters, and more. These updates further enhance the functionality and flexibility of the API.
Remote MCP server support enables developers to manage and monitor multiple servers more easily, improving system scalability and stability. The image generation feature provides users with the ability to generate high-quality images for a variety of fields, such as advertisement design and data visualization.
code interpreterThe addition of the more powerful programming capabilities for developers to execute code directly in the API response, simplifying the development process and improving efficiency.
In addition, these new features support a variety of application scenarios, such as automated testing, data analytics, and the deployment of machine learning models, providing developers with even more possibilities.
This update not only improves the overall performance of the Responses API, but also brings developers more tools and resources to help them achieve greater creativity and productivity in a variety of projects.
---
Developers, please note that the xAI API has just received a major upgrade.
The newly introduced Live Search feature enables Grok to search for data in real time from 𝕏 platforms, the Internet, popular news, and other sources. The introduction of this feature greatly improves the timeliness and accuracy of data acquisition and opens up more possibilities for developers.
Live Search API It is currently in free beta for a limited time. This means that developers can start utilizing this powerful tool for development and innovation immediately.
Whether you're building an intelligent assistant, a data analytics application, or a real-time information retrieval system, the Live Search API provides powerful support. Visit https://t.co/NzSc0vUg7i Start building your project!
The upgrade not only improves the functionality of the API, but also brings more application scenarios for developers. With access to real-time data, developers can respond more flexibly to changing needs and provide more accurate and timely services to users.
---
We are pleased to announce that applications are now open for the Llama Launch Program!
This new initiative is designed to empower early stage startups to help them innovate and build generative AI applications using the Llama platform. By participating in the program, startups will gain the following advantages:
- Cloud service reimbursement to reduce cost burden.
- Hands-on technical support from Llama experts to help you make the most of Llama's features.
- Join a vibrant community and get direct access to a variety of resources to help your startup thrive.
Who can apply?Any registered, early-stage startup located in the U.S. that has raised no more than $10 million and also has at least one developer.
The application deadline is May 30, 2025 at 6:00 p.m. (Pacific Time).
Learn more and apply today!
For more detailed information, please visit:https://t.co/M107ooVWFa
---
We are pleased to announce thatDevstral-- A state-of-the-art (SOTA) open model designed for programming assistants has been released. The model was developed by us and @allhands_ai to improve the efficiency of code generation, debugging and optimization.
Devstral uses the latest deep learning technology to understand complex programming tasks and provide high-quality code suggestions. It not only supports multiple programming languages, but also has powerful contextual understanding and code completion capabilities.
Application scenarios include:
- Automated Code Generation
- Code reviews and quality checks
- Rapid Prototyping
- Developer training and education
By using Devstral, developers can significantly increase productivity, reduce coding errors, and complete complex projects in a short period of time.
Learn more at: https://t.co/LwDJ04zapf
---
Recently, Gemini 2.5 Flash Preview version was officially released, and the biggest highlight of this update is that it realizes the native audio output function through Live API, which makes the dialog interaction more smooth and natural.
This version supports more than 30 different voices, allowing users to choose from a variety of styles and languages based on their needs to build more intuitive and realistic conversational artificial intelligence (AI) agents and experiences. Whether it's customer service, educational tutoring or home assistant applications, this technology will provide users with richer and more personalized interactions.
With advanced text-to-speech (TTS) technology, Gemini 2.5 is able to generate synthesized speech that is close to a real person's voice, further improving the quality and efficiency of human-computer communication. For developers, this means that they now have more tools to create seamless and engaging voice interfaces, driving intelligent conversations in the direction of a more humanized system.
---
🚀 Open Agent Platform (OAP) is an open source, citizen developer-oriented platform for building, prototyping, and deploying intelligent agents. The platform is designed to make it easy for even users without deep programming knowledge to create and manage complex automated workflows.
With OAP, users can:
- 🔧 Build agents through a web interface without complex coding
- 🧠 Connect to RAG servers for more efficient information retrieval
- 🔌 Integration of external tools through MCP extension agents
- 🤖 Orchestrating Multi-Agent Workflows for Complex Task Management with Agent Supervisor
Whether you're a business analyst, product manager, or developer, OAP allows you to leverage the power of LangGraph agents without the need for deep programming knowledge.
---
We encountered many challenges when managing a single codebase containing over 650 community packages. To address these issues, we developed an open source build management tool - theLlamaDevThe
In this post, we'll share our experience migrating from Poetry and Pants to uv and LlamaDev, enabling a faster and simpler development process:
- Test Run Speed Up 20%
- Clearer logs and easier debugging
- Simplify local development with uv integration
- Getting Started Becomes Easier for LlamaIndex Contributors
LlamaDevNot only has it improved our development efficiency, but it has also significantly increased community engagement. By providing clearer logs and a streamlined debugging process, developers can locate and resolve issues faster. In addition, the uv integration makes setting up a local development environment simple and intuitive, significantly reducing the barrier to entry for new contributors.
We believe that these improvements will make our community more active and attract more developers to join.
---
We are pleased to announce that LMArena has successfully raised$100 millionof seed funding to support our platform and continued research on reliable AI. This round was led by leading investment firms and UC Investments, and we are honored to have the support of these investors who believe in our science and mission.
LMArena started as an academic project at UC Berkeley and has grown into a startup valued at $600 million. Our goal is to build a neutral, open and community-driven platform to help globally understand and improve the performance of AI models in real user queries.
In addition, we have big news coming next week!
In order to better meet the needs of our users, we will be re-launching LMArena in the near future, bringing with it a brand new interface design that has been built from the ground up entirely based on community feedback. See the link for more information.
Through this financing, LMArena will further strengthen its technological development and expand its application in education, healthcare and other fields to provide users with more accurate and efficient services.
