Gemini news and informations in 2025

2025 was a massive “agentic” era for Gemini, transitioning from a chatbot into a proactive assistant capable of reasoning and taking action.

Here are the most important news and updates from last year:

The Evolution of Gemini Models

Google moved through two major generations of the model in 2025, focusing on speed and “deep” thinking.

• Gemini 2.0 (Early 2025): Launched in early February, this generation introduced Gemini 2.0 Flash (for speed) and Gemini 2.0 Pro (for complex coding). It featured a massive 2-million-token context window, allowing it to process thousands of pages of text or hours of video at once.+1
• Gemini 2.5 (Mid 2025): Announced at Google I/O in May, this update introduced Deep Think mode. This allows the model to “pause and reason” through multiple hypotheses before answering, significantly improving performance in advanced math and science.+2
• Gemini 3.0 (Late 2025): Released in November, Gemini 3 became the new flagship. It focused on native multimodality, meaning it processes audio, video, and text simultaneously with almost zero latency, making conversations feel truly human.

Major Feature Releases

Google expanded where and how you can use Gemini with several “drops” throughout the year.

• Gemini Live (Now Free): Previously a paid feature, Gemini Live (the conversational voice mode) became free for all iOS and Android users in May 2025. It also added a camera-sharing feature, letting you point your phone at a broken sink or a math problem to get real-time help.
• Project Mariner & Agent Mode: Google introduced “Agents” that can actually use your browser to do things for you—like booking flights, comparing products, or filling out forms—instead of just telling you how to do it.
• Gemini in Chrome: Gemini was integrated directly into the Chrome browser (Windows/macOS), allowing you to summarize any webpage or ask questions about the site you are currently viewing without leaving the tab.

Creative & Professional Tools

• Flow & Veo 3: Google launched Flow, a filmmaking tool powered by the Veo 3 video model. It can generate 4K video clips with native audio (sound effects and speech) directly from text prompts.
• Gemini Canvas: A new workspace for writers and designers that can transform a simple research report into a full website, infographic, or interactive quiz.
• Google AI Ultra: A new $250/year subscription tier was launched, giving “power users” first access to experimental models like Gemini 3 Deep Think and higher limits for video generation.

Integration Highlights

• Android 16 & Pixel 10: Gemini became the “heart” of the new Android OS, handling on-device tasks like real-time translation and “Smart Reply” which learns your specific writing style from Gmail and Drive.
• Google Home: Gemini now writes “event descriptions” for your Nest cameras (e.g., “The cat is playing with a box in the hallway”) and allows you to create complex home automations using natural language.

2026 is truly the year where Gemini has shifted from a conversational partner to an action-oriented agent. Setting these up involves moving beyond just typing prompts and into “configuring” how Gemini interacts with your browser, apps, and code.

Here is a guide to setting up and using the latest agentic features:

1. Project Mariner (Web Agent)

This is the feature that allows Gemini to actually “drive” your Chrome browser to complete tasks like booking flights or researching venues.

• How to Set Up:
  1. Go to labs.google.com/mariner.
  2. Install the Project Mariner extension from the Chrome Web Store.
  3. Pin the extension to your bar for easy access.
• How to Use:
  • Open the extension sidebar while on any website.
  • Give a command like: “Compare the top 3 hotels in Rome for under $200 and draft an email to my partner with the options.”
  • Watch the Live View: You will see the agent opening tabs and clicking buttons. You can click “Take Over” at any point if you want to finish the task manually.

2. Gemini 3 “Deep Think” Mode

If you have a complex problem that requires reasoning (like a business strategy or a difficult coding bug), you can toggle this mode to make Gemini “think” before it speaks.

• How to Activate:
  1. Open the Gemini app or go to gemini.google.com.
  2. Ensure you have Gemini 3 Pro selected in the model picker.
  3. Look for the Deep Think toggle (usually a brain or sparkle icon) in the prompt bar.
• What to Expect: It may take 30–60 seconds to respond because it is evaluating multiple solutions in the background. You can even view a “Thought Summary” to see its internal logic.

3. Gemini Agent (Personal Assistant)

This handles multi-step tasks across your Google Workspace (Gmail, Calendar, Drive).

• How to Enable:
  1. Go to Settings > Extensions in the Gemini app.
  2. Ensure Google Workspace and Personal Intelligence (Beta) are toggled ON.
• Practical Example:
  • “Organize my inbox for the week: prioritize emails from my clients and draft replies for the ones asking for quotes.”
  • The agent will create a list of “Proposed Actions” for you to approve before it sends anything.

4. Agent Mode for Developers

If you use VS Code or Android Studio, the new “Agent Mode” can now write and fix code across multiple files autonomously.

• Setup:
  1. In your IDE, open the Gemini Code Assist chat.
  2. Toggle the Agent switch at the top of the chat window.
• Pro Tip: Create a file named AGENT.md in your project root. Write your coding style and rules there; the agent will read this file every time it performs a task to ensure it follows your specific standards.