Google started rolling out its hotly anticipated Gemini 3 mannequin Tuesday as the corporate goals to remain forward of rivals OpenAI and Anthropic. CEO Sundar Pichai described Gemini 3 as a “state-of-the-art” and “most clever” mannequin, designed to be extra succesful at coding and reasoning via advanced queries.
The corporate mentioned it’s making Gemini 3 out there for everybody to attempt within the Gemini app for Google AI Professional and Extremely subscribers; in AI Mode in Search; for builders via the Gemini API in AI Studio, Google Antigravity, its new agentic improvement platform, and Gemini CLI; and for enterprises in Vertex AI and Gemini Enterprise.
Since launching Gemini two years in the past, Google’s AI chatbot has entered the mainstream. The Gemini app has over 650 million customers, whereas AI Overviews are utilized by 2 billion individuals each month.
What can Gemini 3 do?
Google has highlighted that Gemini 3 is designed for robust multimodal understanding and is a robust mannequin for “vibe coding.” The corporate claims that Gemini 3 outperforms its predecessor on each AI benchmark, topping the LM Enviornment leaderboard with a rating of 1501 factors, which demonstrates PhD-level reasoning, incomes prime marks on Humanity’s Final Examination and GPQA Diamond.
Gemini 3 Professional additionally excelled in multimodal reasoning, scoring 81 per cent on MMMU-Professional and 87.6 per cent on Video-MMMU, showcasing how Google’s frontier mannequin can remedy advanced issues throughout numerous matters, together with science and arithmetic.
Throughout a briefing, Google demonstrated how Gemini 3 understands each visible and spatial data, after which delivers outcomes. For instance, Gemini 3 can decipher and translate handwritten recipes in several languages right into a shareable cookbook. By offering Gemini with tutorial papers and video lectures, it might generate code to create interactive flashcards, making it simpler and extra visually participating to study new matters.
For builders, Google markets Gemini 3 as one of the best mannequin for vibe coding and as its most superior agentic coding mannequin, outperforming the earlier Gemini 2.5 Professional. Builders can construct with Gemini 3 in Google AI Studio, Vertex AI, Gemini CLI, and Google Antigravity, the corporate’s new agentic improvement platform.
Story continues beneath this advert
Google is asking Gemini 3 its most safe AI mannequin, noting that it has undergone intensive security evaluations. “The mannequin reveals diminished sycophancy, elevated resistance to immediate injection, and improved safety in opposition to misuse through cyberattacks,” the corporate mentioned.
Gemini brings a brand new kind of interface to the Gemini app. (Picture credit score: Google)
New interface, Dynamic View
Customers may even discover that the Gemini app has been cleaned up, and redesigned with a extra trendy look. Google has launched what it calls “generative interfaces,” a brand new kind of interface that seems dynamically when prompted. For instance, if you happen to ask Gemini to plan a “three-day journey to Rome subsequent summer season,” you’ll obtain a visible itinerary just like {a magazine} format, full with images and modules.
Google can be including one other layer to the expertise known as Dynamic View, which designs and codes a customized person interface in actual time. As an illustration, if you happen to ask Gemini to “clarify the Van Gogh gallery with life context for each bit,” you’ll get an interactive response that you would be able to faucet and scroll via.
New to the Gemini app is Gemini Agent, powered by Gemini 3. It’s a device that handles multi-step duties straight throughout the app – it might connect with your Google apps, handle your calendar, add reminders, prioritise to-dos, and even draft replies. Gemini Agent is at present rolling out to Gemini AI Extremely customers first.
Story continues beneath this advert
Google can be bringing Gemini 3 to Search, beginning with AI Mode. By deciding on “Pondering” from the mannequin drop-down menu in AI Mode, Gemini 3 permits customers to ask advanced and nuanced questions, together with prolonged ones, permitting the AI to higher perceive the context and intent earlier than responding. Gemini 3 in AI Mode is at present rolling out to all customers within the US, with increased limits out there to these with Google AI Professional and Extremely plans.
Google additionally introduced a brand new agent platform known as “Google Antigravity,” which lets builders code “at the next, task-oriented degree.”
Aggressive AI race
The 12 months has been ruthless in the case of AI dominance, and each main AI firm is overtly difficult its rivals to construct higher and sooner LLMs.
Google is becoming a member of the race, following within the footsteps of OpenAI and Anthropic, which launched their very own cutting-edge fashions simply a few months in the past. It’s unclear whether or not this mixture of velocity, energy, and options might be sufficient, nonetheless. And whereas Gemini 3 is touted as a robust mannequin with robust multimodal capabilities and reasoning expertise, Google nonetheless faces stress from each new and established gamers because the market fills with frontier fashions which might be extra succesful at reasoning and more and more fitted to autonomous device use.
Story continues beneath this advert
The larger query is who will finally dominate the AI market, and whether or not that management will come from the US, China, or someplace else. In January, Chinese language upstart DeepSeek shocked the world with its open supply mannequin, exhibiting a rising wave of highly effective Chinese language fashions now vying for world affect. Later, Mark Zuckerberg’s Meta shaped a brand new superintelligence crew that has lured away a number of prime OpenAI scientists, though no person is aware of precisely what the corporate is attempting to attain.