Google unveils Gemini 3, a new AI system built for deeper reasoning and advanced multimodal understanding (Photo: File)
Google has unveiled Gemini 3, its newest model and the centerpiece of the company's rapid push towards more capable and human-like AI systems.
It launches with great expectations after the success that met Gemini 2.5. Early indications are that Google now has a system that can stand up to and in some ways eclipse, its competitors, like ChatGPT 5 and Grok 4.
According to Sundar Pichai, Gemini 3 is a blend of multimodal intelligence, long-context understanding and an instinctive sense of intent. The objective is simple: to minimize the need for over-prompts and deliver responses to users that can be closer to natural reasoning.
For Pichai, the ability to "read the room" is really what sets this generation apart in product development from pure content processing to nuanced situational understanding.
Demis Hassabis describes Gemini 3 as a significant step toward AGI. The Pro version is designed to handle a range of high-end analytic tasks from scientific problem solving to sophisticated math.
ALSO READ: Last Sunset of 2025: Arctic Slips Into a 64-Day Polar Night With No Sunrise
According to Google, the model scores strongly on demanding benchmarks like Humanity's Last Exam, GPQA Diamond and MathArena Apex-all of which test nuanced reasoning. Hassabis said Gemini 3 doesn't use any filler language or waste words, simply focusing on direct, thoughtful answers, making it feel more deliberate than earlier incarnations.
The new system understands everything from handwritten notes to dense academic papers and full video lectures. Google sees everyday uses including digitizing old family recipes, summarizing long research material or even turning multi-hour videos into clear, structured guides. On Search, Gemini 3 powers a refreshed AI Mode that can create interactive layouts, simulations and tool-driven results.
ALSO READ: Headless Comet? K1 ATLAS Splits Into Four Pieces as It Speeds Toward Earth
Google has also previewed Gemini 3 Deep Think, an extended-reasoning mode meant for the highest difficulty tasks. It reportedly outperforms the Pro model in internal testing and shows promise when it comes to coding, hypothesis checking and tool-use workflows. For now, it is limited to researchers until Google can finalize its safety and performance.
For developers, Gemini 3 introduces expanded coding capabilities and more dependable agentic behaviour. It now tops the WebDev Arena leaderboard and performs well in key tests such as SWE-bench Verified.
Similarly, Google is introducing Antigravity, a workspace that allows AI agents to work across editor, terminal and browser environments, letting them plan and perform full workflows while keeping the user in control.
ALSO READ: Google Launches ‘Gemini 2.5 Computer Use’: AI That Can Click, Type, and Scroll Like Humans | Key Features
Disclaimer: This article is based on available announcements and testing data, features may evolve as Google updates Gemini 3 and its related tools.
