What's new in Gemini

Gemini · Google|Web #2 · Mobile #3

1. New featureNovember 12, 2026Official · Google Gemini products

   Google's Gemini Now Available on Apple CarPlay for Hands-Free Voice Chats

   Google's AI app Gemini is now available on Apple CarPlay, which connects an iPhone to a car's display. Drivers can now hold voice conversations without touching the screen.

   • Gemini Live, a real-time voice chat feature, now works on CarPlay.
   • Users can ask it to manage their calendar, summarize emails, or brainstorm ideas by voice.
   • Chats are saved to the iPhone, so users can pick them up again after they park.
   • To use it, update the Gemini app on the iPhone and turn on microphone permission.

   Read moreSource

2. New featureNovember 11, 2026Official · google-gemini/gemini-cli releases

   Gemini CLI v0.68.0 Released With Mostly Bug Fixes

   Google's Gemini CLI, a command-line tool for using Gemini, has a new version, v0.68.0. This update focuses mainly on fixing bugs rather than adding new features.

   • Windows file path checks and IDE process lookup were fixed.
   • Chat history no longer redraws excessively after the terminal window is resized.
   • The loading screen now shows rate-limit (429) wait times and a countdown.
   • Extension installs running at the same time were ordered to avoid conflicts.

   Read moreSource

3. New modelNovember 11, 2026Official · Google Gemini products

   Google Launches MKA Benchmark for Visually Complex Knowledge Work

   Google DeepMind, working with Artificial Analysis, released MKA (Multimodal Knowledge Agent), a benchmark that tests how well AI agents (AI that carries out multi-step tasks on its own) handle work where visuals and layout matter, such as blueprints, schematics, reports, and spreadsheets. It includes 200 tasks written by domain experts across 10 fields and 83 occupations.

   • Each task is a multi-step job that takes a skilled professional three to eight hours and is graded with about 70 yes/no checks.
   • Scoring weights factual accuracy at 75% and visual design at 25%.
   • Grading is done by an AI from a non-Google company, which matched expert grades 96% of the time.
   • Ten public tasks are on Hugging Face, while the 200-task set stays private.

   Read moreSource

4. New featureNovember 10, 2026Official · Google Gemini products

   Google brings Adobe PDF and design tools into everyday Google apps

   Google said it is expanding its partnership with Adobe so people can use Adobe's document and design tools inside Google apps like Drive, Gmail, and Gemini. The goal is to let people finish their work without switching between apps.

   • Users will soon be able to go straight from Google Drive and Gmail into Adobe Acrobat, a PDF editing program.
   • Adobe's creative tools will also be available in Gemini and in AI Mode in Google Search.
   • Businesses will be able to use an Adobe AI assistant in Gemini Enterprise to help with design, PDF work, and marketing campaigns.
   • Adobe and Google DeepMind will explore near-real-time editing features, starting with an image model for Photoshop.

   Read moreSource

5. New featureNovember 10, 2026Official · Google AI blog

   Google Cloud Launches Gemini for Healthcare in Preview

   Google Cloud has launched Gemini for healthcare, an AI tool that helps hospitals and insurers with paperwork. It pulls together scattered information such as medical records, billing, and clinical guidelines to handle multi-step tasks. It is currently available in preview.

   • It handles multi-step jobs such as summarizing medical records and reviewing prior authorization requests.
   • The AI works within set safety rules, and people check its work.
   • It is first being tested with partners including HCA Healthcare, Highmark Health, and Northwell Health.

   Read moreSource

6. New modelNovember 10, 2026Official · GitHub Changelog

   Google's Gemini 4 Carbon Now Generally Available in GitHub Copilot

   Google's latest AI model, Gemini 4 Carbon, is now generally available in GitHub Copilot. It is built for complex coding and security-focused tasks.

   • Copilot Pro+, Max, Business, and Enterprise users can pick it in the model selector.
   • It works in tools including Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode, and Eclipse.
   • The rollout is gradual, so users who don't see it yet should check back later.
   • Administrators of Business and Enterprise plans can decide whether to allow the model through the model policy.

   Read moreSource

7. New modelNovember 10, 2026Official · Google DeepMind blog

   Google DeepMind unveils Gemini Robotics 2.5 to speed up robot learning

   Google DeepMind has introduced Gemini Robotics 2.5, a new AI model that helps robots learn more tasks faster in real-world settings. It is a VLA (vision-language-action) model, meaning it understands sight and language and turns them into physical actions. It is currently rolling out to robotics partners first.

   • The model was tested on more than 200 tasks, including serving food with tongs, tying a bow on a gift box, and pipetting liquid.
   • Compared with earlier models, its success rate is 30% to 100% higher depending on the task, and it finishes tasks 20% to 30% faster.
   • In a Mercedes-Benz production line test, a humanoid robot sorted and packed car parts with a 96% success rate.
   • The team improved physical safety and created a new benchmark that tests robots against adversarial jailbreak attempts.

   Read moreSource

8. Pricing & limitsNovember 6, 2026Press · Ars Technica

   Google Says Schools Can Skip Gemini but Must Keep AI Search Enabled

   Google's new AI policy for elementary, middle, and high schools has drawn growing protests from teachers and parents. Google says schools will be able to choose for themselves whether to use the AI features.

   Read moreSource

9. New modelNovember 6, 2026Official · Google Gemini products

   Google Launches Gemini Omni 1.2 Flash, Extending Videos to 15 Seconds and Cutting Prices 30%

   Google has released Gemini Omni 1.2 Flash, a video generation AI model. It can create videos up to 15 seconds long at once, and its price is 30% lower than the previous version. It is available right away in the Gemini API, Google Vids, and Gemini for enterprise.

   • Video length has increased from 10 seconds to 15 seconds. Users can choose any length from 3 to 15 seconds in 1-second steps.
   • Users can revise videos through multiple back-and-forth conversations. They can edit only parts without starting over from scratch.
   • The people, products, and clothing in reference images stay more consistent throughout the video.
   • Pricing is set at $0.07 per second for 720p video. Users can first create a quick low-resolution version and upgrade only the ones they like to high resolution.

   Read moreSource

10. New featureNovember 5, 2026Official · google-gemini/gemini-cli releases

    Gemini CLI 0.67.0 Released with IDE Connection and Model List Command

    Version 0.67.0 of Gemini CLI, Google's AI tool for use in the terminal, has been released. This version includes several new features and many bug fixes.

    • It can now connect to an IDE (code editor) and be used inside a container sandbox.
    • A command to view the list of available models and a --list-models option have been added.
    • Several bugs have been fixed, including issues with settings migration, running tools after a command is canceled, and character display width.
    • An error for exceeding the project spending limit is now classified as a quota error at the termination level.

    Read moreSource

11. New modelNovember 3, 2026Official · Google AI blog

    Google's October AI Roundup: Gemini 4 Argon and Carbon Launch

    Here is a summary of Google's main AI announcements from October 2026. It first released Gemini 4 Argon, which is focused on reasoning (the process where AI works through steps to reach an answer). Two weeks later, it announced the more powerful Gemini 4 Carbon. It also showed Genie 4, which creates virtual worlds that move in real time.

    • Gemini 4 Argon has been released worldwide for complex tasks such as coding, financial analysis, and legal work.
    • Gemini 4 Carbon is still being tested only by the U.S. government and a small group of trusted security experts. It will be released widely after its safeguards are refined.
    • Genie 4 lets users steer virtual worlds with realistic movement and sound by voice, and it can be used to train robots and AI agents.
    • Google has expanded its SynthID Detector, which checks whether images, videos, or sounds were made by AI, to a worldwide English-language audience.

    Read moreSource

12. New featureOctober 30, 2026Official · Google Gemini products

    Gemini Now Lets You Create, Preview, and Edit Docs, PDFs, Slides, and Spreadsheets in Chat

    Google's Gemini app now lets you create and preview files such as documents, PDFs, presentations, spreadsheets, and code right inside the chat window. You can also expand the view to full screen and edit the text directly. The feature is available in every language where Gemini is offered, to all users.

    • You can create and preview files directly in the chat window.
    • You can drag to select text and then ask questions about it or ask Gemini to revise that part.
    • PDFs can be edited on the web and on mobile, while Google Docs, Sheets, and Slides can only be edited on the web.
    • You can upload handwritten lecture notes and ask Gemini to turn them into PDF study materials.

    Read moreSource

13. New featureOctober 29, 2026Official · Google Gemini products

    Google's Gemini App Adds Immersive Learning Library for Exploring Topics

    Google's Gemini app now includes a collection of immersive explorations, visual learning content where users tap through layers of a topic. Users can learn topics ranging from the microscopic world to outer space by tapping directly on the content. The collection is available first in the Student Hub.

    • Topics include the Silk Road trade routes, the inside of a jet engine, and the physics of lightning.
    • Tapping a label takes users deeper, from planet-scale views down to molecules and atoms.
    • An audio narration feature (guided explanations delivered by sound) that explains each stage is expected to be added soon.
    • Users will also be able to access these explorations directly within chat conversations, according to the company.

    Read moreSource

14. New featureOctober 29, 2026Official · Google Gemini products

    Google and XREAL Open Orders for AURA, a Wired XR Glasses Device

    Google and XREAL have launched AURA, wired XR glasses, and started taking orders in the United States, Canada, and South Korea. Japan can place orders starting November 2. The glasses show app screens in front of your eyes while you still see your surroundings.

    • The lenses are transparent, so you can see your surroundings while using virtual screens.
    • A portable unit called the compute puck handles the heavy processing and connects to the glasses by cable, keeping the glasses under 95 grams.
    • Google's AI assistant Gemini can see the apps on screen and your surroundings and help you in real time.
    • By connecting a laptop with a USB-C cable, you can display the laptop screen as a window in your view.

    Read moreSource

15. New featureOctober 29, 2026Official · Google Gemini products

    Google Previews Platform for Building and Managing Connected Apps in One Place

    Google has started a preview of a Connected Apps platform that lets developers build and manage apps used in Search and the Gemini app in one place. Partner apps such as Adobe, Cam, and Peloton are also being added.

    • Developers connect apps using an MCP server (a channel that helps AI connect with external tools) hosted in the cloud, then register and test them in the Google Console before submitting.
    • Before review, developers can choose whether to make the app available in Search, the Gemini app, or both.
    • After launch, developers can check usage, response delays, and error rates, and manage logos and account authentication.
    • So far, more than 50 connected apps have been released across Search and Gemini, and the Cam integration will soon be usable in both Search and Gemini.

    Read moreSource

16. New modelOctober 28, 2026Official · Google DeepMind blog

    Google DeepMind Unveils Genie 4, a Real-Time World Simulation Model

    Google DeepMind has released Genie 4, a new world model. World models are AI systems that create virtual environments. The company says it shows much more realistic scenes in real-time simulations than earlier versions.

    Read moreSource

17. New modelOctober 28, 2026Official · Google Gemini products

    Google Shares Eight Examples of What Gemini 4 Argon Can Do

    Google has shared eight ways people have used its new AI model, Gemini 4 Argon. The company says it can finish complex tasks that used to take a long time in just a few minutes. The examples include research organization, 3D graphics, games, and presentation design.

    • It built a 3D library scene from scratch using math code and also created a video of the process.
    • It made a 3D game track that can be controlled, starting from a photo of a desk.
    • It organized more than 80,000 original NASA files and turned them into a presentation for executives.
    • Where it can be used depends on payment. Developers and businesses can access it through the API, while Google AI Ultra and Pro subscribers can use it in the app and in AI Mode in Search.

    Read moreSource

18. New modelOctober 24, 2026Official · Google Gemini products

    Five Builders Show What They Created With Google's Gemini 4 Argon Model

    Google has released Gemini 4 Argon, a new AI model that it says is good at complex tasks such as coding, math, and building 3D scenes. Google showcased five works made by developers.

    • One developer created a 3D scene of an aircraft over water, including light, reflections, and ripples.
    • Another pulled earthquake and wildfire data to build a 3D globe that moves in real time.
    • A jet engine was built as an interactive 3D model whose parts can be separated and moved.
    • A 3D carriage was created from a single photo, with working doors, lanterns, and wheels that react to bumps.

    Read moreSource

19. New modelOctober 14, 2026Official · Google DeepMind blog

    Google Begins Safe Public Rollout of Gemini 4 Argon With Safety Safeguards

    Google DeepMind has started releasing its new AI model, Gemini 4 Argon, broadly to many fields. It was first given only to trusted cyber defense teams, and during that period its effectiveness was confirmed, including finding a critical flaw in hospital software. The company also explained how it prevents misuse.

    • It solved all 12 cyberattack scenarios, up sharply from 5 for the previous model.
    • Its ability to refuse risky requests improved over the previous model, and violations in adversarial testing fell by almost half.
    • It has three layers of real-time protection, including input checks and internal signal monitoring, to try to block bypass attempts.
    • A detailed safety report has been released separately.

    Read moreSource

20. New modelOctober 14, 2026Official · Google AI blog

    Google Releases Gemini 4 Argon, Claiming Top Rankings for Complex Professional Tasks

    Google has released its latest AI model, Gemini 4 Argon. Developers, businesses, scientists, and general users can use it, and the company says it excels at complex real-world work. It also said the model ranked first on benchmarks across several fields.

    • It handles multi-step tasks across many fields, including coding, finance, law, cybersecurity, and science.
    • API pricing is an introductory rate of $2 per million input tokens and $10 per million output tokens.
    • Google's internal development teams used it for large code cleanup work, and agent code submissions rose by more than 35%.
    • It was used to make xTB, an open-source quantum chemistry tool, up to 2.3 times faster.

    Read moreSource

21. New featureOctober 7, 2026Official · Google AI blog

    Google Unveils Playground, a Tool That Turns Text Descriptions Into Playable Games

    Google has released Playground, an experimental platform that lets people make games just by describing them in writing, even if they have no coding experience. People can share the games they make with friends or post them to a community gallery. It runs directly in a web browser.

    • You can describe an idea in writing, and the platform can build the characters, rules, and world.
    • Finished games can be kept private, sent to friends through a link, or posted to a public gallery.
    • It supports multiplayer games and leaderboards (scoreboards that show rankings).
    • Unity Spark (a game development engine, a type of professional game tool) is planned to be added later.

    Read moreSource

22. New modelOctober 7, 2026Official · Google DeepMind blog

    Google Releases EmbeddingGemma 2, a Lightweight Multimodal Model That Runs On Devices

    Google DeepMind has released EmbeddingGemma 2, a lightweight AI model that handles text, images, audio, and video together. It can run embedding directly on devices such as smartphones. Embedding means turning data into a list of numbers so that similar items are easier to find. The model is released under the Apache 2.0 license.

    • The previous version only handled text, but this version can group and search code, images, video, and sound in one shared space.
    • It has about 740 million parameters (the learned values in an AI model), and it is designed to run well on devices.
    • With text alone, it works with only 270 million parameters, and image and sound features can be added separately.
    • The company says it can be used to find a specific video scene from a voice memo, or to search several hours of recordings by text.

    Read moreSource

23. New modelOctober 3, 2026Official · Google AI blog

    Google Recaps Its September 2026 AI Announcements, Including Gemini 4 Argon

    Google has summarized its main AI announcements from September 2026. It unveiled Gemini 4 Argon, a high-performance reasoning model, and added more voice models and app integrations. In science, it highlighted a human DNA mapping effort and a case of tracking methane in space.

    • Gemini 4 Argon is said to be strong at solving complex problems and is designed especially for cybersecurity defense.
    • Argon can produce results for up to 1 million tokens (units of text) at once.
    • Argon will first be released to trusted security experts, then expanded to developers and general users after feedback.
    • Gemini 3.8 Flash is said to deliver better performance at the same speed and cost as 3.7 Flash.

    Read moreSource

24. New featureOctober 1, 2026Official · Google DeepMind 블로그

    AI Tool PlanAssist Cuts Building Permit Review Time by 58%

    PlanAssist, an AI tool built by Google DeepMind, helps government officials review building permit applications, such as requests to expand homes. In a pilot program, the time officials spent reviewing each case dropped by an average of 58%. Final decisions are still made by officials.

    • PlanAssist checks whether all required application documents have been submitted and whether the plan follows cultural heritage protection rules and local and national planning policies.
    • It also checks whether residents or others have raised problems or objections in the past.
    • It writes a draft review report with sources, which officials then revise and use to make the final decision.
    • In more than 70% of the pilot program reports, officials approved the draft as it was, with no changes needed.

    Read moreSource

25. New featureSeptember 1, 2026Official · Google DeepMind 블로그

    Google DeepMind Unveils SynthID Bio, a New Tool Announced Without Details

    Google DeepMind has introduced something called SynthID Bio. Beyond the name, no further details are available.

    Read moreSource

AI reads official posts and trade press and rewrites them in plain words. Some items may be missing or wrong; check the source for details.