Gemini 3 launches: Heres every new feature and how to try them
Google has officially unveiled Gemini 3, its latest and most sophisticated AI model to date, which the company proudly proclaims as its “most intelligent” system yet. This launch marks a significant milestone, as it is the first time a new Gemini release has been integrated into Google Search from day one. The introduction of Gemini 3 is part of Google’s ongoing commitment to advancing AI technology, focusing on creating systems that are not only more capable but also more context-aware. In a recent announcement, Google highlighted that this model showcases deeper reasoning abilities, enhanced multimodal understanding, and a greater awareness of user intent compared to its predecessors.
One of the standout features of Gemini 3 is its advanced reasoning capabilities. Google asserts that this model is designed to understand nuance and tackle complex problems more effectively than earlier versions. In benchmarking tests, Gemini 3 Pro achieved an impressive score of 1501 on the LMArena leaderboard, outperforming competitors like Grok and Gemini 2.5 Pro. Additionally, it showed remarkable improvements in academic-style reasoning, scoring 37.5% on Humanity’s Last Exam and 91.9% on GPQA Diamond without the aid of external tools. The introduction of an enhanced mode called Gemini 3 Deep Think further elevates these capabilities, achieving even higher scores on various benchmarks, including a notable 41% on Humanity’s Last Exam and 45.1% on the ARC-AGI-2 challenge, making it ideal for tackling the most challenging long-horizon reasoning tasks.
Gemini 3 is already operational across the Google ecosystem, with features available in multiple products. Users can access the AI Mode in Search, which offers dynamic generative interfaces and complex reasoning support, although this is currently limited to Google AI Pro and Ultra subscribers. The Gemini app has also been redesigned to enhance user experience, featuring a new “My Stuff” folder and improved shopping functionalities. Developers can experiment with Gemini 3 through AI Studio, Vertex AI, and the Gemini CLI, while enterprise customers have the option to deploy the model via Vertex AI and Gemini Enterprise. For those eager to experience Gemini 3 firsthand, immediate access is available through the AI Mode in Search (subscription required) or by using the Gemini app, where Gemini 3 is now the default model. The Deep Think version, which promises even more advanced capabilities, is set to roll out in the coming weeks, following additional safety evaluations. This launch positions Google at the forefront of AI advancements, promising enhanced user experiences across its platforms.
Google has officially launched Gemini 3
, its newest and most advanced AI model — and the company is calling it its “most intelligent” system yet. The model rolled out across Google products on day one, marking the first time a new Gemini release has gone live in Search at launch.
Gemini 3 represents what Google execs describe as the next major step in their push toward more capable, more context-aware AI.
In a note sharing the launch
, the company stated that the model incorporates deeper reasoning, stronger multimodal understanding, and enhanced awareness of user intent compared to earlier versions.
What’s new in Gemini 3?
The headline upgrade is reasoning. Google says Gemini 3 is built to understand nuance, break down complex problems, and “grasp depth and context” in ways its predecessors couldn’t. In benchmarking tests, Gemini 3 Pro took the top spot on the LMArena leaderboard with a 1501 Elo score, outperforming both Grok and Gemini 2.5 Pro.
SEE ALSO:
Google’s Gemini 3 is smarter, better at figuring out what you need
The model also demonstrates significant improvements in academic-style reasoning. On Humanity’s Last Exam, Gemini 3 Pro hit 37.5 percent without any external tools, and on GPQA Diamond, it reached 91.9 percent. It also posted state-of-the-art scores across math and multimodal tasks, including video understanding and factual accuracy.
A new enhanced mode called Gemini 3 Deep Think pushes those capabilities even further. Deep Think improves scores on nearly every benchmark, including a jump to 41 percent on Humanity’s Last Exam and 45.1 percent on the ARC-AGI-2 challenge. It’s designed for the most challenging, long-horizon reasoning tasks.
Where can you use Gemini 3?
Gemini 3 is already live across the Google ecosystem. That includes:
AI Mode in Search
, with more dynamic, generative interfaces and complex reasoning support. (Currently available for Google AI Pro and Ultra subscribers.)
The Gemini app
is getting a redesign, including the My Stuff folder and improved shopping experiences.
Gemini for developers
, available now in AI Studio, Vertex AI, Gemini CLI, and Google’s new agentic development platform Antigravity.
Gemini Agent
, an upgraded system that can complete multi-step tasks like inbox organization, appointment management, and more.
How to try Gemini 3 right now
If you want hands-on time with Gemini 3 immediately, there are a few pathways:
Use AI Mode in Search
(Google AI Pro or Ultra subscription required).
Open the Gemini app
, where Gemini 3 is the default model starting today for most users globally.
Developers
can experiment with AI Studio or Vertex AI, gaining full access to Gemini 3 Pro.
Enterprise customers
can deploy it through Vertex AI and Gemini Enterprise.
Deep Think, the highest-end version, will be rolled out in the coming weeks following additional safety evaluations.
