Google has officially launched its Gemini 3 Pro model, and for the first time, it's already making its way directly to the public, without making you wait months after the announcement to actually get your hands on it. On top of introducing a few new features, Google says the model has the typical increased accuracy, but also, that it finally cuts down on some of the excessive people-pleasing flattery that drives me nuts when using AI. The catch? Most features are paywalled, and one is still in the oven.For the more obvious improvements, Google says Gemini 3 Pro is at the top of LMArena with a score of 1501 points, while also demonstrating "PhD-level reasoning" on Humanity's Last Exam, earning a 37.5% score without the use of tools. Math buffs will also be happy to hear that the new model scored 91.9% on GPQA Diamond and 23.4% on MathArena Apex.But for everyone else, the real excitement comes from the brand new things you can do in Gemini 3 Pro. This isn't just a performance bump. Instead of just being Gemini 2.5 but better, Gemini 3 Pro also debuts three new consumer-facing features, plus a new platform for developers. It's a much meatier release than, say, ChatGPT 5.1—assuming you're willing to pay up.Generative UIGoogle's press release bounces back and forth between calling this feature "Generative UI" and "Generative Interfaces," although I prefer the former. Essentially, it's supposed to make your AI results easier to read.It's one of the few new features that's freely open to everyone, although Google has two different approaches to it, and not everyone might see the same one. Credit: Google The first is called "visual layout," and is more similar to Gemini's current results pages. Essentially, when you enter a complex, multi-layered prompt, like "plan a three-day trip to Rome next summer," you'll now get an explorable visual itinerary, rather than static text. This will include photos and clickable modules, but remain within the Gemini interface you've gotten used to. It may also have interactive sliders and buttons for further refining your search, but Google says the idea is to give you an "immersive, magazine-style view." Credit: Google The second, then, is more like an on-demand webpage. It's called "dynamic view," and basically uses agentic coding to generate an on-the-fly app to help you learn more about a topic. This will include generated text and imagery, but may appear quite different from the Gemini interface you've gotten used to. An example in the press release sees Google generating a dynamic view response to "explain the Van Gogh Gallery with life context for each piece," which creates a scrollable page with a clickable header, the art justified on the left, and scrollable text on the right with clickable subheadings and pull quotes, all with custom font and design that can differ pretty wildly from Gemini's other output.One of the issues I've had with lengthy plain-text AI responses is that they can get a bit tiresome to skim, and Google's hoping this will help deal with that. Still, it hasn't settled on an approach. Both visual layout and dynamic view are rolling out today, to free and paying customers, but the company says that "to help us compare these experiments, you may initially see only one of them."Gemini 3 Deep ThinkAs a contrast to Generative UI, Gemini 3 Deep Think is behind a hefty paywall, and still in the oven. The feature is an evolution on the existing Gemini 2.5 Deep Think mode, and essentially allows the AI to take more time to answer a question so it can better reason out an appropriate response. It's similar to existing free deep research modes across Gemini and other AI apps, but more broadly applicable, and Google says it's great for use cases like intricate graphic design or coding.Unfortunately, it's only in the hands of a few "safety testers" for now. Google told press it will start to release "in the coming weeks," but even then, it'll be limited to Google AI Ultra subscribers. Since Google AI Ultra is $250/month (although it starts at $125/month for the first three months), that's a pricey proposition. Gemini AgentAnother Google AI Ultra-exclusive feature, Gemini Agent will start rolling out to subscribers today, and is similar to the company's recent AI shopping push. The idea is for it to take action for you, to help you handle multi-step tasks without having to leave Gemini. Credit: Google The key thing is that it works with Google's other apps. So, for instance, you could ask it to help organize your Gmail inbox, and it'll separate out your pending emails into categories and submit emails it thinks you could delete for your approval. Or, you could ask it to help you rent a car under a certain budget for an upcoming trip, and it'll scan your emails for flight and hotel details, then search for an appropriate booking and reach out to you before finalizing it. It's got access to the web, Google Workspace, and other AI tools like Canvas, so in theory, it can pull from pretty much every resource Google has to help you answer your question. Google also says it will "seek confirmation before critical actions like making purchases or sending messages, and you can take over any time."If it works, it sounds like the type of virtual secretary most people probably thought of when these companies started talking about AI just a few years ago. But with such a high monthly price tag, it's probably only for the highest of rollers for now. So, kind of like a regular secretary, I guess.Google AntigravityFinally, this feature is intended more for developers than the average internet user, but it's worth bringing up, if only because it's free. Called Google Antigravity, it's a new development platform focusing on agentic coding, AKA having the AI generate code for you. It's more complicated than that—you're able to freely browse and edit generated code—but the idea is to make it easier to use Gemini as a development partner. Credit: Google It's not Gemini's first development tool, but the idea is to give developers a dedicated AI workspace. To that end, Antigravity can pull from existing features like Canvas, as well as implement "browser control capabilities" and "asynchronous interaction patterns" to achieve an "agent-first product form factor" that can "autonomously plan and execute complex, end-to-end software tasks."I'm sure the people who would use Antigravity will know what all that means. To a layperson like me, it seems like the big improvement is that it's one app that you can use to go straight from ideation to publishing, rather than having to bounce between several different Gemini tools.What's probably more interesting, though, is that Antigravity is free, "with generous rate limits on Gemini 3 Pro usage." Yep, developing with Gemini is actually now cheaper than using Gemini, depending on what you want to do.How to use Gemini 3At least some of Gemini 3 Pro is available to everyone. To try Gemini 3 Pro for yourself, open the Gemini app or webpage and select "Thinking" from the model selector underneath your prompt. Google told me free users will have up to five Gemini 3 Pro prompts per day, while AI Plus, Pro, and Ultra subscribers will "enjoy higher limits."Google AI Pro and Ultra subscribers will also be able to use Gemini 3 Pro right from Google search's AI Mode, also by selecting "Thinking" from the model selector (next to the "AI Mode" button). Free users will have to stick to the Gemini app for now, unfortunately, but Google says that will change "soon." On the plus side, AI Mode will still be able to generate visual layouts and dynamic views, assuming you have access to Gemini 3 Pro within it.AI Overviews will also start to use Gemini 3 Pro for AI Pro and Ultra subscribers, although that's set for "the coming weeks." When it arrives, AI Mode will also be upgraded to automatically send your hardest questions to Gemini 3 Pro, without you having to pick it in the model selector (although you can continue to manually pick older models if you prefer).Personally, I'm glad to see Google releasing Gemini 3 Pro alongside concrete new features, instead of just making "AI, but better." At the same time, because most of this requires a subscription, it's clear we're still a little while away from that wide AI adoption the tech industry still seems to be clamoring for.