Google has set the bar high for Gemini 3. It’s promising a bunch of advanced features in its shiny new AI models, from code that generates interactive 3D visualizations to “agent” capabilities that complete tasks. But as we have seen in the past, what is advertised does not always match reality. So we tested some of Google’s claims and found that the Gemini 3 performs quite well, with a few caveats.
Google announced the Gemini 3 family of models earlier this week, with the flagship Gemini 3 Pro being the first to be introduced to users. The Gemini 3 Pro is believed to come with major upgrades in logic, as well as the ability to provide more concise and direct responses than Google’s previous models.
Some of the biggest promised improvements are in Canvas, the built-in workspace inside Gemini apps, where you can ask the AI chatbot to generate code, as well as preview the output. Building in Canvas, Google says Gemini 3 can interpret content from a variety of sources, such as text, images, and video at the same time. According to Google, the model can also handle more complex signals, allowing it to generate richer, more interactive user interfaces, models, and simulations. The company says Gemini 3 is also “exceptional” in the zero-shot generation, meaning it’s better at completing tasks it hasn’t been trained for.
For my first test, I tried one of the more complex requests that Google showed in one of its demos: I asked Gemini 3 to create a 3D visualization of the differences in scale between a subatomic particle, an atom, a DNA strand, a beach ball, the Earth, the Sun, and a galaxy, as shown here.
Gemini 3 created an interactive visual similar to the one shown by Google, allowing me to scroll through and compare the sizes of different elements, which seemed to correctly list each one from smallest to largest, starting with the proton and maxing out all the way to the cosmic web. (To be fair, I would Hope Gemini can detect that a beach ball is much smaller than the Sun.) It included almost everything shown in the demo, but its image quality diminished in some areas, as the 3D models of the DNA and the beach ball’s strands were significantly dimmer than those shown by Google. I saw much the same thing when feeding other Google demos into Gemini. The model conveyed the correct concept, but it was always a little shoddy, whether it had lower resolution or was simply a little more cluttered.
When I tried something simple the Gemini 3’s output didn’t match Google’s demo. I asked it to recreate a model of a voxel-art eagle perched on a tree branch, and while my results were similar to the demo, I couldn’t help but notice that the eagle had no eyes, and the trees were trunkless. From Google’s example, the voxel-style panda turned out fine, but the standard 3D models of the penguin and turtle turned out to be very primitive, with little or no detail.
But the Gemini 3 isn’t just made for prototyping and modeling; Google is testing a new “Generative UI” feature for Pro subscribers that packages their responses inside a “visual” magazine-style interface, or as a “dynamic” interactive webpage. I only got access to Gemini 3’s visual layout, which Google showed off as a way to visualize your travel plans, like a three-day trip to Rome.
When I tried out the Rome trip prompt, the Gemini 3 presented me with an itinerary that looked like a personalized webpage, along with options to further customize it, such as whether I’d prefer a leisurely or fast-paced vacation or whether I’d prefer certain dining styles. Once you submit your preferences, Gemini 3 will redesign the layout to match your selection. I found that this feature could also provide interactive guides on other topics, like how to build a computer or set up an aquarium.
Next, I did a little experiment with the Gemini Agent, a feature that Google is testing for Ultra subscribers inside the Gemini app. Like other agentic features, Gemini Agent aims to perform tasks on your behalf, such as adding reminders to your calendar and creating reservations.
An example shared by Google shows the Gemini Agent organizing a Gmail inbox, so I asked the tool to do the same — and, well, it followed my orders. It found 99 unread emails from the past week and displayed them inside an interactive chart. Gemini provided options to set reminders for the most important things like RSVPs and bills, while also offering buttons to archive emails identified as promotional. I asked Google Gemini to schedule a reminder to pay my bill, and the AI assistant put it into a Google Task with the correct due date. When I asked it to pay the bill, it navigated the billing interface and came close to asking me to enter my payment details, but (given security concerns around agent AI) I stopped it from proceeding.
Although you can organize your inbox manually, I found the Gemini 3’s assistance somewhat helpful, as it found some forgotten emails that I might have missed. You can also tell Gemini to find and unsubscribe from a large number of spam email providers, which is nice.
Between Perplexity’s AI assistants, ChatGPT, and Gemini, Google’s AI chatbot (presumably) offers the richest integration with Gmail. Tangle will pull the listed emails into your inbox, but you’ll only have to tell it which ones to keep, archive, or delete rather than pressing a button, as you can do with Gemini. For some reason, ChatGPT refused to organize my inbox, claiming that its integration with Gmail was in “read-only” mode, despite me easily sending an email through the app on my behalf. But while Gemini is directly connected to Gmail, it was still much slower at sending emails in the app than Perplexity.
Gemini About Managed to book a restaurant reservation without any intervention, only to incorrectly tell me there was a “cost” associated with doing it right before the booking was finalized. When I asked about the fee, Gemini 3 pushed back and said it “probably refers” to the restaurant’s 16 percent service charge. It then asked me to confirm my reservation three times And then told me that there was again a financial transaction involved. breathThen, I realized I could complete these tasks myself much faster,
Despite the obstacles in completing tasks, the Gemini 3 Pro’s interactive visualization features were impressive, and I could see how interactive models or visual layouts could be useful in some scenarios – although I can’t see myself using them on a daily basis, and Gemini’s text-based answers are generally informative enough for me. For now, I think I’ll continue to use Gemini as I always have: for queries that I can’t immediately find when browsing the web.