Google’s Gemini 3 is finally here and it’s smarter, faster, and free to access

The new model is also fueling a new agentic coding assistant.

Table of Contents

Gemini 3 Pro: What it can do Safety A deeper dive into performance How to access (including an app redesign)

ZDNET’s key takeaways

Google finally launched its Gemini 3 model.
The model is faster, smarter, and better at coding and multimodality.
It also powers a new agentic coding app and even Search.

With its Gemini AI models, Google has integrated its AI assistant across several of its most popular platforms, including Search, Google Workspace, and Android devices. Now, the company is releasing its latest and greatest model to upgrade user experiences throughout its ecosystem.

On Tuesday, Google finally launched Gemini 3, which the company claims is the “best model in the world for multimodal understanding and our most powerful agentic and vibe coding model yet.” The claim is supported by benchmark data, crowd-sourced Arena results, and more advanced use cases that the chatbot has not previously been able to tackle.

Also: Let Gemini make your next slide presentation for you – here’s how

The best part? Everyone can try it across Google’s suite of tools, including Search, the Gemini app, AI Studio, and Vertex AI, as well as a brand-new agentic development platform, Google Antigravity.

To learn about what’s new and what you can expect, keep reading below.

Gemini 3 Pro: What it can do

Google is kicking off the Gemini 3 era with the launch of Gemini 3 Pro. Gemini 3 Pro outperforms its predecessor, Gemini 2.5 Pro, across all major AI benchmarks — but before I get into the data, here’s what it practically means for you.

Gemini 3 Pro has been designed to provide users with better quality answers that have a new “level of depth and nuance,” according to Google. This includes more smart, concise, and direct answers that focus on helpfulness rather than flattery.

Of course, as the benchmarks support, the model is designed to be smarter and reason more effectively, demonstrating improved performance in Ph.D.-level reasoning that enables it to better tackle complex problems, such as those in science and mathematics.

Also: How to turn off Gemini in your Gmail, Photos, Chrome, and more – it’s easy to opt out of AI

Google said the model also pushes the frontiers of multimodal reasoning by being able to synthesize information about a topic across multiple modalities, including text and images, video, audio, and code. For everyday help, this means you can input information in more ways, have the model understand, and output responses in whichever multimodal medium is best fit for your response.

Lastly, for developers, the model reflects a major win, as it’s the best vibe coding and agentic coding model Google has released, even climbing to the top of the WebDev Arena, a leaderboard compiled through user votes after comparing the anonymous models’ ability to build a site and voting for the best one.

Google

Developers can leverage coding capabilities through Google AI Studio, Vertex AI, Gemini CLI, and third-party platforms such as Cursor, GitHub, JetBrains, and others. The company also launched Google Antigravity, a brand-new agent development platform that functions as an active coding partner, allowing agents to plan and execute complex, end-to-end software tasks and even validate their own code.

Also: Gemini is gaining fast on ChatGPT in one particular way, according to new data

The agentic capabilities don’t end there, with Google sharing that it will be able to take action on your behalf with more complex workflows, such as reorganizing your Gmail inbox or booking services. However, these more advanced capabilities are made accessible through the Gemini Agent in the Gemini app for Google AI Ultra subscribers.

Safety

Alongside performance upgrades, Google said that Gemini 3 is its most secure model yet and has undergone the most comprehensive safety evaluations of all models to date. Some of the company’s findings include “reduced sycophancy, increased resistance to prompt injections, and improved protection against misuse via cyberattacks.”

If you are interested in seeing results from assessments, you can review the Gemini 3 model card.

A deeper dive into performance

Gemini 3 Pro skyrocketed to the top of the LMArena Leaderboard with a score of 1,501 points, a noteworthy feat as that score is compiled through anonymous voting against top models from Anthropic, Meta, XAI, DeepSeek, and more.

Reflecting its advanced reasoning, the model scored a 91.9% on the GPQA Diamond, a benchmark used to evaluate Ph.D.-level reasoning, a “Top-score” on Humanity’s Last Exam (37.5% without usage of tools), a multimodal exam that tests a variety of fields, and an 81% on the MMMU-Pro and 87.6% on the Video MMMU, benchmarks used to test multimodal reasoning.

Google

Also: Gemini for Home is finally rolling out for early access – here’s how to try it first

The advanced reasoning capabilities also translate to higher scores on Math benchmarks, such as the Math-Arena-Apex, where the model scored a state-of-the-art 23.4%, according to Google.

The company also launched Gemini 3 Deep Think, which has deeper intelligence, reasoning, and understanding capabilities. The Deep Think version outperformed Gemini 3 on Humanity’s Last Exam (41% without using tools) and GPQA Diamond (93.8%).

How to access (including an app redesign)

Everyone can access Gemini 3 through the Gemini app, which has been redesigned to feature a clean, modern look, according to Google.

The app redesign also makes it easier to find the content you have created, included images, videos, and reports in the “My Stuff” folder. The shopping experience was also enhanced to now include “product listings, comparison tables, and prices from Google’s Shopping Graph.”

Also: I let Gemini Deep Research dig through my Gmail and Drive – here’s what it uncovered

Google has also added two new “generative interfaces,” visual layout and dynamic view, which present information in different ways depending on your preference and the task you are doing. These are part of an experiment and are being rolled out to users today. As part of the experiment, you may just see one of them.

To access Gemini 3 in AI Mode in Search, users must have a Google AI Pro or Ultra subscription. Developers can access it in the Gemini API in AI Studio, Antigravity, and Gemini CLI.

Lastly, enterprises can access it in Vertex AI in Gemini Enterprise.

Follow ZDNET: Add us as a preferred source on Google.

From zdnet.com

Categories:

uncategorised