Gemini Release Notes

Pixel Halloween Features

J Jack O. Lantern CSO - Chief Scare Officer

This Halloween season, you may already be conjuring the perfect costume, or researching your trick-or-treating route. Now, get ready to summon a terrifyingly good tool to go along with it: Pixel. Here are a few frighteningly fun features to try, including the latest additions released with the new Pixel 10 phones, the Pixel 10 Pro Fold and the Pixel Watch 4.

• Try a festive Pixel Watch face. In the Pixel Watch app, select “watch faces” and then the edit icon in the upper righthand corner. You’ll see an option for a photo watch face. Once you select it, go to Google Photos and select up to 30 photos that your watch will show on its display. Try scrolling back to past Octobers for some nostalgic Halloween picks, right on your wrist.

• Get the entire group in the shot. Speaking of group shots: You might be getting together with friends and family quite a bit this month, whether that’s to visit the pumpkin patch, swap soups, trick-or-treat, watch scary movies or hit up a costume party. Pixel 10’s Auto Best Take will make sure you all look your best in your group pic, and you can use Add Me to get everyone — even the photographer — in there. And if you have a Pixel 10 Pro Fold, you can rely on Instant View to immediately show you if you got the shot: One half of the phone will show you the photo while you can continue to shoot via the live feed on the other half — that way, you’ll really know if you got the perfect group selfie or need to take a few more.

• Want a spooky shot? Use Camera Coach. Another helpful feature for making sure you get the image you want is Camera Coach. Maybe you’re carving pumpkins and need help figuring out how to frame them all or you want to capture an artsy image of someone’s costume: Camera Coach will ask what you want to accomplish with your photo, then guide you through to the result.

• Generate frightening or festive photos. There are so many ways you can use Pixel for some scary fun with photos: You can ask Gemini to turn a photo from last Halloween into a short video or to make your pic look like it’s from the ancient past. There’s also Pixel Studio, where you can use AI to create any spooky scene you can imagine. And there’s no shortage of ways to play with images in Google Photos — like asking it to put your pup in a costume or creating a fall-themed collage.

• Record an eerie track. Pixel’s Recorder app now lets you add background music to a recording. Hosting a Halloween party or a haunted house? You can record some eerie sounds and set background music to them. You can choose to create your own vibe (the “angry” and “sad” options work best for scary party purposes and would easily complement ghostly “ooOOooOOoo”s!) and Pixel will generate a track for you.

  You can record some eerie sounds and set background music to them.

• Adjust to darker days. Halloween’s approach isn’t the only sign of fall: There are also the darker, shorter days. Help yourself adjust to the season change with Pixel’s Sunrise Alarm, which gradually brightens your screen over 15 minutes before your audio alarm goes off. Even if you can’t actually wake up with the sun these days, this can help getting up in the dark feel a bit more natural.

POSTED IN:

• Pixel
• Photos
• Gemini App

Original source Report a problem

Gemini can now help you schedule time with others in Gmail.

Today we announced a new Gemini-powered feature in Gmail called "Help me schedule". Here's how it works:

When you're responding to an email about finding a time to meet, Gmail detects the intent and shows a Help me schedule button in the toolbar. Select it, and Gemini suggests meeting times based on your Google Calendar and the context of the email. The time suggestions are added directly to your reply, so the other person can choose what works best.

Once they pick a time, a Calendar invite is automatically created for everyone. Learn more about this update on the Workspace Updates blog.

Original source Report a problem

Helping students get the most out of Gemini

AI can broaden and expand access to knowledge in powerful ways, helping anyone, anywhere learn anything in the way that works best for them. It’s not about just getting an answer, but deepening understanding and building critical thinking skills along the way. That’s why we built

Guided Learning

, a new mode in Gemini that acts as a learning companion guiding you with questions and step-by-step support. With Guided Learning, students can work through things like complex math problems, structure arguments, get started on an essay, prep for a test, get homework help, test their understanding with interactive quizzes, and more. Students are also exploring new uses for AI, unlocking more creativity and finding fun, helpful ways to integrate it in their everyday lives. With our new, best-in-class image generation and editing model

Nano Banana

, students can reimagine photos and turn abstract ideas into striking, practical visuals. They can quickly mock up concepts like their ideal dorm room design, prototype a logo for a campus club, design a whiteboard concept for a group project, or virtually try on outfits for a presentation. With Veo 3, students can transform text and images into high-fidelity, cinematic clips, accelerating creative work and allowing students to quickly mock up scenes for a film project or generate b-roll footage for a presentation.

Veo 3

: Transform text or a photo into a

8-second video

with sound

• Nano Banana: Transform images in amazing new ways with higher access on 2.5 Pro
• 2 TB of storage : tons of space for notes, projects, photos and papers on Google Photos, Drive and Gmail

Learn more and sign up for the 12-month Google AI Pro plan, at no cost, by December 9th.

Expanding access to AI-powered tools for educators

We recently introduced

Gemini for Education

, a version of the Gemini app built for the unique needs of the educational community, and made

Gemini in Classroom

available to all Google Workspace for Education editions free of charge with

more than 30 new capabilities

to help teachers plan more efficiently and create engaging content and resources. We’re also

partnering with universities

around the world to help educators and students use tools like Gemini and NotebookLM to enhance efficiency, get more personalized support and foster AI literacy. By giving students access to our most powerful tools, we’re equipping them with AI skills to boost creativity, spark curiosity, and enhance problem-solving skills. We can’t wait to see what the next generation of creators, artists and thinkers will do.

Original source Report a problem

Video Overviews now look even better thanks to Nano Banana, Gemini’s latest image generation update. Plus, try a new "Brief" format for quick insights.

Eugene Lo Software Engineer, NotebookLM

Dense documents can be challenging. NotebookLM helps you understand any sources you upload and today, that process becomes better — and more fun, too.

A creative boost with Nano Banana

Nano Banana in NotebookLM generates helpful, contextual, and beautiful illustrations based on the sources you upload. The result is Video Overviews that don't just tell you about your documents — they help you understand and remember them.

New Video Overviews will automatically use one of six new visual styles: Watercolor, Papercraft, Anime, Whiteboard, Retro Print and Heritage.

Two ways to watch

Sometimes you need a detailed summary from a Video Overview. Other times, you just need the highlights. NotebookLM now offers two formats to meet both needs.

• Explainer: A structured, comprehensive video based on your sources for in-depth understanding.
• Brief: A new, bite-sized video format to help you quickly grasp the core ideas of a document.

To create the new video overview:

1. Select your sources in NotebookLM and click the "Video Overview" button.
2. To customize your video, select the pencil icon on the Video Overview tile.
3. Choose between the format, visual style and try customizing the video, like "Focus only on the cost analysis sections of the business plan." or "Convert these recipes into an easy-to-follow video focusing on the prep time and cooking steps."
4. Sit back and explore your notebook while the video is being generated.

This update continues our work to make information more accessible and useful by transforming dense information into dynamic multimedia that helps people understand complex topics in new ways.

The new Video Overviews will start rolling out in all supported languages to Pro users this week and to all users in the upcoming weeks.

Original source Report a problem

Gemini Enterprise

We’re introducing Gemini Enterprise, designed to bring the full power of Google’s AI to every employee, for every workflow.

Editor’s note:

Today, our CEO Sundar Pichai will join Cloud CEO Thomas Kurian at our Gemini at Work event on Google’s campus to announce Gemini Enterprise.

It’s been an exciting year for Google Cloud: we surpassed a $50 billion annual revenue run rate in Q2, with 13 product lines each generating over $1 billion in annual revenue. Much of this growth is driven by AI, with 65% of our Cloud customers already using our AI products. Those customers include nine of the top ten AI labs and nearly all AI unicorns.

AI has opened up new opportunities in the cloud market — and today, we announced our next chapter with Gemini Enterprise.

Beyond simple chatbots Gemini Enterprise is designed on the premise that true business transformation in the era of AI must go beyond simple chatbots. You need a comprehensive and integrated platform that brings all your company's data, tools, and people together in one secure place.

That’s exactly what we’ve built. Gemini Enterprise is an AI-powered conversational platform designed to bring the full power of Google AI to every employee for every workflow.

Built with Google's most advanced Gemini models, it enables you to chat with your company’s documents, data, and applications. It also gives you the tools to build and deploy AI agents, as well as a suite of pre-built agents, and is grounded in your company’s information and your personal context at work.

We think of it as the new front door for AI in the workplace. And we're already seeing how customers are using AI tools in powerful ways:

• HCA Healthcare: To improve the critical handoff of patient information during shift changes, HCA Healthcare is piloting a Gemini-powered Nurse Handoff solution, built in collaboration with nurses and grounded in best practices. The automated handoff report is always reviewed by nurses for accuracy before they end their shift, and is estimated to save millions of hours annually.
• Best Buy has transformed its customer service, driving a 200% increase in customers rescheduling deliveries on their own and resolving 30% more questions on topics like price matching and recycling.
• We’re using AI here at Google too, across everything from operations to coding to marketing to supply chain management. Today, nearly half of all new code at Google is generated by AI, reviewed and accepted by engineers. This helps our engineers do more and move faster.

Our full-stack approach to AI Gemini Enterprise is the latest example of how our full-stack approach to AI enables us to innovate faster for our Cloud customers.

• AI infrastructure: It starts with our extensive and reliable infrastructure, which powers all of Google’s products, including Google Search and YouTube. This includes GPUs from our partner Nvidia, and our purpose-built Tensor Processing Units (TPUs). Our latest generation, Ironwood, offers a 10x performance improvement over the previous generation and will be generally available soon.
• Research: Our world-class research teams Google Research and Google DeepMind push the frontiers of science, robotics, health and autonomous driving. For the second year in a row our team member received a Nobel Prize. Michel Devoret, our chief scientist on our quantum team, was honored with the Nobel in Physics for research he did alongside John Martinis and John Clarke in the 1980s, and has paved the way for our quantum efforts today. He joins the ranks with Demis Hassabis and John Jumper, who were awarded a Nobel in Chemistry for their breakthrough AlphaFold last year. AlphaFold is one example of Google DeepMind’s pioneering work which includes breakthroughs like AlphaGenome, Gemini robotics, and state-of-the-art weather prediction.
• Models: Our foundation models, led by Gemini, introduce new capabilities with each generation. Gemini 2.5 Pro has topped LMArena for over 6 months, and as of this writing is still number 1 across the Text and Vision leaderboards. Our model accuracy is a key reason why healthcare organizations like Seattle Children’s Hospital and ASCO choose Google Cloud. Our generative AI model portfolio is the most extensive in the world, including Veo, Imagen and Nano Banana for image editing. Over 13 million developers have built with our generative models, including Gemini, optimizing for performance, speed and cost.
• Products and platforms: This all comes to life in our products and platforms, bringing AI to more people than anyone else. AI is enabling hugely popular features across Google Workspace, AI Overviews in Search are used by over 2 billion people, and this summer, we reached a milestone of 1.3 quadrillion monthly tokens processed across our surfaces (up from 980 trillion monthly tokens announced in July).

Gemini Enterprise is the entry point for organizations to access our full-stack innovations, and we’re excited to see how it helps teams do even more with AI.

You can read even more from Google Cloud CEO Thomas Kurian on the Cloud blog.

Original source Report a problem

Gemini Enterprise works across all the tools and data you use every day to make your workflows easier — giving you time back in your day.

Maryam Gholami Senior Director, Product Management

AI is presenting a once-in-a-generation opportunity to transform how we work. But the first wave of AI, while promising, has been stuck in silos, unable to connect the dots across an entire organization.

Today, we are changing that by introducing Gemini Enterprise: the new front door for AI in the workplace. It brings the best of Google AI to every employee, moving beyond simple tasks to transform entire workflows and how we work. It’s an advanced platform that does something fundamentally new: it brings your work, data, and people together in one AI-powered hub.

Instead of leaving you to stitch different tools together, Gemini Enterprise provides a single, secure environment where any employee can easily build and deploy AI agents to automate processes.

Here’s how Gemini Enterprise can help you and your team get time back in your day.

1. Give every team an AI specialist.

Imagine giving every team a specialist assistant with expert-level skills on demand. Gemini Enterprise makes this possible. With a powerful, no-code workbench, any user — from marketing to finance — can build custom AI "agents" to handle their most tedious tasks.

For example, a marketing manager can kickstart a new campaign in minutes instead of days. They can ask an agent to generate creative concepts for a new product launch. Because the agent is securely connected to the company's data, it instantly understands the product's details and has access to all the approved creative assets, like logos, brand and product imagery. The agent can then generate a variety of high-quality, on-brand campaign ideas—complete with social media copy and visual mockups—ready for the team to review. This allows teams to go from a simple prompt to polished, actionable ideas at a scale that was previously impossible.

You also don't have to start from scratch. To deliver value from day one, Gemini Enterprise includes a number of pre-built Google agents, like our Deep Research and Data Science agents. And the platform is built on a principle of openness. Through the agent marketplace, you can discover, filter, and deploy thousands of vetted agents from our rich ecosystem of partners, integrating powerful third-party solutions directly into your organization's workflows.

2. Connect the dots across all your business data.

Your company's most valuable information is often trapped—hidden away in different spreadsheets, applications, and chat threads. Gemini Enterprise is designed to unlock that knowledge, letting your teams interact with it in a secure way, and through a simple and intuitive chat interface.

By securely connecting to the systems you use every day, from Google Workspace and Microsoft 365 to core business applications like Salesforce and SAP, Gemini Enterprise goes beyond simple search to build context. This gives your AI agents the business context they need to deliver truly intelligent, relevant results.

For example, by giving an agent context from their internal analytics and business intelligence systems, relationship managers at Banco BV can automate hours of work, freeing them up to connect with customers. Similarly, Harvey uses Gemini to provide legal AI with deep contextual understanding, making lawyers far more efficient in contract analysis and compliance, saving them hours of time.

3. Automate entire workflows, not just tasks.

True transformation comes from improving entire processes, not just single steps. That’s why Gemini Enterprise is designed to orchestrate complex workflows across the different systems your teams rely on every day.

By combining pre-built, custom, and third-party agents, you can automate and orchestrate entire processes on your behalf. Macquarie Bank, for instance, uses Google Cloud AI to enable more efficient and proactive digital self-service capabilities. Their Help Centre Search now directs 38% more users towards self-service and they have reduced false positive alerts for client protection by 40%.

4. Use powerful AI in the tools you love.

The best AI shouldn't force you to switch contexts; it should meet you where you are. That's why Gemini Enterprise is designed to work seamlessly across the platforms your teams rely on, including both Google Workspace and Microsoft 365.

While Gemini Enterprise is a powerful standalone platform, it is deeply integrated with Google Workspace, unlocking further benefits. This is where you can experience our first multi-modal agents that understand and create text, images, video, and speech, built right into the apps your teams already use:

• Video: With Google Vids, you can transform a presentation into an engaging video, complete with an AI-generated script and voiceover.
• Voice: In Google Meet, we're bringing real-time speech translation to business customers, capturing tone and expression to make conversations seamless across languages.

Get started today

This is about more than making one task easier. It's about making entire workflows smarter and giving people time back to focus on high-impact work. By bringing all these components together through a single interface, Gemini Enterprise transforms how teams work.

Ready to transform how your team works? Learn more about Gemini Enterprise.

Original source Report a problem

Here’s a recap of some of our biggest AI updates from September, including an AI upgrade for Chrome, a new visual search option in AI Mode and the next step in bringing helpful robots into the physical world.

For more than 20 years, we’ve invested in machine learning and AI research, tools and infrastructure to build products that make everyday life better for more people. Teams across Google are working on ways to unlock AI’s benefits in fields as wide-ranging as healthcare, crisis response and education. To keep you posted on our progress, we're doing a regular roundup of Google's most recent AI news.

Here’s a look back at some of our AI announcements from September.

Forget pumpkin spice; the real news from September was the massive stack of AI updates. AI delivered major updates across our most popular core services like Chrome, Search and Android, making them significantly smarter. And the Gemini app became a powerhouse with its latest Gemini Drop, featuring the viral Nano Banana, Gemini Live, custom and shareable Gems, and the no-code app-building tool, Canvas. Meanwhile, even as we integrate AI across our digital products, Google DeepMind is busy working to bring helpful robots into the physical world.

With AI our overarching goal remains the same: to make AI as useful as possible, whether through the fun, visible features that help you with everyday tasks or in the essential, behind-the-scenes magic boosting your cybersecurity and learning.

Chrome gets smarter with AI

We added major new AI features to Chrome. Gemini in Chrome now acts as an AI browsing assistant, letting you answer questions and find information across all of your open tabs. We also introduced AI Mode in the omnibox for asking complex, multi-part questions, alongside future agentic capabilities that will automate multi-step tasks like ordering groceries. Plus, AI is now keeping you safer by proactively blocking new types of scams and enhancing security and privacy features. To take a deeper look at the update, we shared how AI was built into the new, shinier Chrome.

Search gets more powerful with AI

We upgraded AI Mode in Search, making it easier to get inspired and search visually. By combining Gemini 2.5 and our new "visual search fan-out" technique, we’ve unlocked a deeper understanding of images and your natural language questions. Now, you’ve got stunningly precise visual search results that make everything from shopping to exploring new room designs more intuitive than ever before.

We shared five tips for Search Live, a new way to get help in real-time. By integrating an interactive voice conversation in AI Mode with the ability to share your phone’s camera feed, we’ve created a new way to get multimodal help in real-time. Search can now literally see what your camera sees and respond instantly, providing hands-free help with tasks like travel exploration, complex troubleshooting and bringing school projects to life.

We expanded AI Mode to new languages. This update brings our most powerful AI search experience, powered by a custom version of Gemini, to new languages globally: Hindi, Indonesian, Japanese, Korean, Brazilian Portuguese and Spanish. The expansion focuses on a nuanced understanding of local information so users can ask complex questions and explore the web more deeply in their preferred language.

The Gemini app becomes a hub for AI

We shared 10 ways you can use Nano Banana in the Gemini app. Since launching in August, Google DeepMind’s image generation and editing model, fondly known as Nano Banana, has quickly grown in popularity in the Gemini app. So we created 10 examples to show how capable and fun the model is, whether you’re looking for more straightforward tasks like swapping outfits in a photo, or interested in complex, imaginative image generation, like showing your adult self having a tea party with a younger you.

We made collaboration easier in the Gemini app with the ability to share custom Gems. Gems allow you to tailor Gemini for specific needs, and now you can share the ones you create with friends, family or coworkers. The sharing process is similar to Google Drive, giving you control over who can view or edit your personalized AI tools like detailed vacation guides or even custom meal planners.

Android gets more assistive with AI

We launched new Android features to help you polish and share what you write. The latest features in Android include new AI writing tools in Gboard to revise your tone and automatically fix spelling and grammar right on your device. We also announced the ability to let two people listen to the same audio simultaneously, introduced a way to create private QR code audio broadcasts and redesigned Quick Share for instant file transfer previews and live progress updates.

Robots become more helpful with AI

We've introduced the next step in bringing helpful robots into the physical world. Google DeepMind is leveling up robotics with Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, kicking off the era of physical agents. These models let robots see, plan, think and use tools to tackle complicated, multi-step tasks far better than before, and allow for learnings to transfer between different types of robots. Think of Gemini Robotics-ER 1.5 as the smart brain, handling the big-picture reasoning (and even Googling stuff!) while Gemini Robotics 1.5 is the mover, turning visual information and instructions into motor commands for a robot to perform a task.

Learning becomes more tailored with AI

We introduced new features into NotebookLM to help with learning. Our latest updates turn NotebookLM into your ultimate personal AI study partner, with a focus on active learning. You can instantly create flashcards and quizzes grounded in your own notes, generate upgraded reports with suggested formats like a blog post or study guide, and try the Learning Guide option for personalized, step-by-step tutoring. Plus, you can now hear your sources in new ways with Audio Overviews that offer perspectives like a Critique or a Debate.

We launched new resources to promote AI literacy for parents, students and educators. These resources include a new podcast for parents called "Raising kids in the age of AI," expanded student programs like the Be Internet Awesome AI Literacy curriculum and the AI Quests game-based experience. This work includes substantial support for teachers, with over 650,000 educators trained so far and $40 million in grants dedicated to scaling AI literacy programs.

We introduced Guided Learning, a new, interactive study partner in the Gemini app. Powered by LearnLM, Guided Learning fine tunes our AI models for education, allowing people to navigate any topic, step-by-step, to ask questions and promote understanding. With helpful videos and images, the result is a personalized tutor that can break down complex code, create study plans from your uploaded material, and guide you to homework solutions without doing the work for you.

Sundar Pichai spoke at the White House AI Education Taskforce. Sundar highlighted Google’s major push to support AI education across the U.S., including offering Gemini for Education to every high school in America. It builds on Google’s broader $1 billion commitment to support AI education in the U.S., including giving all students and teachers access to our best AI tools, putting $150 million towards grants for AI education and digital wellbeing, and expanding our AI for Education Accelerator from 100 to 200 colleges and universities.

And lastly, a gold medal goes to Gemini

We sent our AI to the International Collegiate Programming Contest World Finals. Gemini 2.5 Deep Think achieved a major AI milestone by achieving gold-medal level performance at the International Collegiate Programming Contest (ICPC) World Finals. This breakthrough performance in abstract problem-solving builds on our previous gold at the International Mathematical Olympiad, proving Gemini's world-class coding and reasoning capabilities.

Original source Report a problem

Google AI Plus is coming to 36 more countries.

Google AI Plus is our new plan designed to help people do more with the latest Google AI models and features, for less. After a positive reception in Indonesia and our first wave of 40 countries, we’re expanding Google AI Plus to 36 more countries today.

Google AI Plus includes higher limits for our image generation and editing model, Nano Banana, and more access to video generation in the Gemini app, Flow and Whisk. You also get Gemini built into apps like Gmail and Docs, expanded access to NotebookLM, 200 GB of storage and more.

You can sign up for Google AI Plus today in one of the 77 countries where it is now available. For a limited time, new subscribers get 50% off for the first six months.

Price varies by country.

Original source Report a problem

Google DeepMind

Available in preview via the API, our Computer Use model is a specialized model built on Gemini 2.5 Pro’s capabilities to power agents that can interact with user interfaces.

Earlier this year, we mentioned that we're bringing computer use capabilities to developers via the Gemini API. Today, we are releasing the Gemini 2.5 Computer Use model, our new specialized model built on Gemini 2.5 Pro’s visual understanding and reasoning capabilities that powers agents capable of interacting with user interfaces (UIs). It outperforms leading alternatives on multiple web and mobile control benchmarks, all with lower latency. Developers can access these capabilities via the Gemini API in Google AI Studio and Vertex AI.

While AI models can interface with software through structured APIs, many digital tasks still require direct interaction with graphical user interfaces, for example, filling and submitting forms. To complete these tasks, agents must navigate web pages and applications just as humans do: by clicking, typing and scrolling. The ability to natively fill out forms, manipulate interactive elements like dropdowns and filters, and operate behind logins is a crucial next step in building powerful, general-purpose agents.

How it works

The model’s core capabilities are exposed through the new computer_use tool in the Gemini API and should be operated within a loop. Inputs to the tool are the user request, screenshot of the environment, and a history of recent actions. The input can also specify whether to exclude functions from the full list of supported UI actions or specify additional custom functions to include.

The model then analyzes these inputs and generates a response, typically a function call representing one of the UI actions such as clicking or typing. This response may also contain a request for an end user confirmation, which is required for certain actions such as making a purchase. The client-side code then executes the received action.

After the action is executed, a new screenshot of the GUI and the current URL are sent back to the Computer Use model as a function response restarting the loop. This iterative process continues until the task is complete, an error occurs or the interaction is terminated by a safety response or user decision.

The Gemini 2.5 Computer Use model is primarily optimized for web browsers, but also demonstrates strong promise for mobile UI control tasks. It is not yet optimized for desktop OS-level control.

Check out a few demos below to see the model in action (shown here at 3X speed).

Prompt: “From https://tinyurl.com/pet-care-signup, get all details for any pet with a California residency and add them as a guest in my spa CRM at https://pet-luxe-spa.web.app/. Then, set up a follow up visit appointment with the specialist Anima Lavar for October 10th anytime after 8am. The reason for the visit is the same as their requested treatment.”

Prompt: “My art club brainstormed tasks ahead of our fair. The board is chaotic and I need your help organizing the tasks into some categories I created. Go to sticky-note-jam.web.app and ensure notes are clearly in the right sections. Drag them there if not.”

How it performs

The Gemini 2.5 Computer Use model demonstrates strong performance on multiple web and mobile control benchmarks. The table below includes results from self-reported numbers, evaluations run by Browserbase and evaluations we ran ourselves. Evaluation details are available in the Gemini 2.5 Computer Use evaluation info and in Browserbase’s blog post. Unless otherwise indicated, scores shown are for computer use tools exposed via API.

The model offers leading quality for browser control at the lowest latency, as measured by performance on the Browserbase harness for Online-Mind2Web.

How we approached safety

We believe that the only way to build agents that will benefit everyone is to be responsible from the start. AI agents that control computers introduce unique risks, including intentional misuse by users, unexpected model behavior, and prompt injections and scams in the web environment. Thus, it is critical to implement safety guardrails with care.

We have trained safety features directly into the model to address these three key risks (described in the Gemini 2.5 Computer Use System Card).

Further, we also provide developers with safety controls, which empower developers to prevent the model from auto-completing potentially high-risk or harmful actions. Examples of these actions include harming a system's integrity, compromising security, bypassing CAPTCHAs, or controlling medical devices. The controls:

• Per-step safety service: An out-of-model, inference-time safety service that assesses each action the model proposes before it’s executed.
• System instructions: Developers can further specify that the agent either refuses or asks for user confirmation before it takes specific kinds of high-stakes actions. (Example in documentation).

Additional recommendations for developers on safety measures and best practices can be found in our documentation. While these safeguards are designed to reduce risk, we urge all developers to thoroughly test their systems before launch.

How early testers have used it

Google teams have already deployed the model to production for use cases including UI testing, which can make software development signficantly faster. Versions of this model have also been powering Project Mariner, the Firebase Testing Agent, and some agentic capabilities in AI Mode in Search.

Users from our early access program have also been testing the model to power personal assistants, workflow automation, and UI testing, and have seen strong results. In their own words: “A lot of our workflows require interacting with interfaces meant for humans where speed is especially important. Gemini 2.5 Computer Use is far ahead of the competition, often being 50% faster and better than the next best solutions we’ve considered.” - Poke.com, a proactive AI assistant in iMessage, WhatsApp and SMS with multiple third-party and agentic workflows. “Our agents run fully autonomously, performing work where small mistakes in collecting and parsing data are unacceptable. Gemini 2.5 Computer Use outperformed other models at reliably parsing context in complex cases, increasing performance by up to 18% on our hardest evals.” — Autotab, a drop-in AI agent. “When conventional scripts encounter failures, the model assesses the current screen state and autonomously ascertains the required actions to complete the workflow. This implementation now successfully rehabilitates over 60% of executions (which used to take multiple days to fix).” — Google’s payments platform team, which implemented the Computer Use model as a contingency mechanism to address fragile end-to-end UI tests that contributed to 25% of all test failures.

How to get started

Starting today, the model is available in public preview, accessible via the Gemini API on Google AI Studio and Vertex AI.

• Try it now: In a demo environment hosted by Browserbase.
• Start building: Dive into our reference and documentation (see Vertex AI docs for enterprise use) to learn how to build your own agent loop locally with Playwright or in a cloud VM with Browserbase.
• Join the community: We’re excited to see what you build. Share feedback and help guide our roadmap in our Developer Forum.

POSTED IN: Google DeepMind Gemini Models AI

Original source Report a problem