Google is launching Gemini 3, its ‘most intelligent’ AI model yet
The flagship Gemini 3 Pro model is coming to the Gemini app and Search, with improvements across coding, reasoning, and less ‘flattery.’
Google is officially rolling out Gemini 3 today — a new family of AI models the company calls its “smartest” and most “factually reliable” so far. It’s also a big chance for Google to pull ahead of OpenAI after GPT-5’s shaky release, and maybe even reclaim the spotlight in consumer AI.
For the first time, Google is giving everyone access to its newest flagship model — Gemini 3 Pro — right inside the Gemini app from day one. Subscribers using Google Search will also get the upgraded model. Tulsee Doshi, senior director at Google DeepMind, says this update brings Google a big step closer to its long-held mission of making knowledge “universally accessible and useful,” especially as Search continues to evolve.
She adds, “One of the biggest changes is moving beyond plain text answers and giving you something richer — something you can actually see and experience.”
Gemini 3 Pro is natively multimodal, meaning it can understand text, images, and audio all together, instead of treating them like separate tasks. Google says the model could take photos of your favorite recipes and turn them into a personalized cookbook, or convert a set of video lectures into interactive study flashcards. It’s designed to feel more intuitive — almost like you’re talking to a creative partner instead of a tool.
You’ll notice these upgrades across Google products. In the Gemini app, the built-in workspace called Canvas lets you build deeper, more complete projects. Google is also testing “generative interfaces” in Gemini Labs — a feature that lets Gemini 3 Pro create visual magazine-style layouts, custom UI designs, or interactive pages you can actually browse. It’s the kind of feature that makes you stop and think, “Oh wow, AI can actually do this?”
In Search’s AI Mode, you’ll see more visual responses too — images, tables, grids, even small simulations — depending on what you ask. Google has also upgraded its “query fan-out” system. Now it not only breaks your question into smaller parts, but also better understands what you’re really trying to find, helping it surface content it might have overlooked before.
Google also couldn’t resist taking a light jab at OpenAI. The company says Gemini 3 Pro is far less likely to give the overly flattering, empty-praise style responses that ChatGPT is sometimes known for. According to Doshi, the difference will be “noticeable.” Google describes the model’s tone as “smart, concise, and direct — offering genuine insight instead of clichés or flattery. It tells you what you need to hear, not just what you want to hear.” In short, more honesty, less buttering you up.
Gemini 3 Pro also comes with stronger reasoning and “agentic” abilities, letting it take on more complex tasks and plan ahead with more reliability. These powers fuel an experimental Gemini Agent feature inside the Gemini app, which can handle tasks for you — reorganizing your emails, researching trips, even booking the travel itself. It’s designed to feel like a helpful assistant that’s actually capable, not overwhelmed.
The model now ranks at the very top of LMArena’s AI benchmark leaderboard. A special Deep Think mode boosts its reasoning even further, though for now, only safety testers have access.
Starting today, Gemini 3 Pro is available to everyone through the Gemini app. In the US, Google AI Pro and Ultra subscribers can also try Gemini 3 Pro in AI Mode by choosing “Thinking” from the model menu. Gemini Agent will roll out first to AI Ultra subscribers.
