Google set the bar excessive for Gemini 3. It’s promising a bunch of upgraded options in its shiny new AI mannequin, from producing code that produces interactive 3D visualizations to “agentic” capabilities that full duties. However as we’ve seen in the past, what’s marketed doesn’t all the time match as much as actuality. So we put a few of Google’s claims to the check and located that Gemini 3 delivers moderately nicely — with caveats.
Google introduced the Gemini 3 household of fashions earlier this week, with the flagship Gemini 3 Professional rolling out to customers first. Gemini 3 Professional is meant to return with large upgrades to reasoning, together with the power to supply extra concise and direct responses in comparison with Google’s earlier fashions.
A few of the greatest promised enhancements are to Canvas, the built-in workspace inside the Gemini app, the place you’ll be able to ask the AI chatbot to generate code, in addition to preview the output. When constructing in Canvas, Google says Gemini 3 can interpret materials from completely different sorts of sources on the identical time, like textual content, pictures, and movies. The mannequin can deal with extra complicated prompts as nicely, permitting it to generate richer, extra interactive person interfaces, fashions, and simulations, in line with Google. The corporate says Gemini 3 is “distinctive” at zero-shot generation, too, which implies it’s higher at finishing duties that it hasn’t been educated on.
For my first check, I attempted out one of many extra complicated requests that Google confirmed off in one among its demos: I requested Gemini 3 to create a 3D visualization of the distinction in scale between a subatomic particle, an atom, a DNA strand, a seaside ball, the Earth, the Solar, and the galaxy, as shown here.
Gemini 3 created an interactive visible much like what Google demonstrated, permitting me to scroll by means of and evaluate the scale of various parts, which appeared to appropriately checklist each from small to massive, beginning on the proton and maxing out on the cosmic internet. (To be truthful, I’d hope Gemini might determine {that a} seaside ball is far smaller than the Solar.) It included nearly all the things proven within the demo, however its picture high quality fell quick in a few areas, because the 3D fashions of the strand of DNA and seaside ball had been fairly dim in comparison with what Google confirmed. I noticed a lot of the identical when feeding Google’s other demos into Gemini. The mannequin spit out the proper idea, but it surely was all the time just a little shoddier, whether or not it had decrease decision or was just a bit extra disorganized.
Gemini 3’s output didn’t fairly stack as much as Google’s demo after I tried one thing just a little easier, both. I requested it to re-create a mannequin of a voxel-art eagle sitting on a tree department, and whereas my outcomes had been fairly much like the demo, I couldn’t assist however discover that the eagle didn’t have any eyes, and the timber had been trunkless. Branching out from Google’s instance, a voxel-style panda got here out alright, however customary 3D fashions of a penguin and turtle got here out fairly primitive, with little to no element.
However Gemini 3 isn’t simply constructed for prototyping and modeling; Google is testing a brand new “generative UI” characteristic for Professional subscribers that packages its responses inside a “visible” magazine-style interface, or within the type of a “dynamic” interactive webpage. I solely received entry to Gemini 3’s visible structure, which Google confirmed off as a method to envision your journey plans, like a three-day journey to Rome.
Once I tried out the Rome journey immediate, Gemini 3 offered me with what seemed like a personalised webpage that includes an itinerary, together with choices to customise it additional, akin to whether or not I’d favor a relaxed or fast-paced trip or if I prioritize sure eating types. When you submit your preferences, Gemini 3 will redesign the structure to match your choices. I discovered that this characteristic can present interactive guides on different matters, too, like the best way to construct a pc or arrange an aquarium.
Subsequent up, I did just a little experimenting with Gemini Agent, a characteristic Google is testing for Extremely subscribers contained in the Gemini app. Like different agentic options, Gemini Agent goals to carry out duties in your behalf, akin to including reminders to your calendar and creating reservations.
One instance shared by Google exhibits Gemini Agent organizing a Gmail inbox, so I requested the software to do the identical — and, nicely, it adopted my orders. It discovered 99 unread emails from the final week and displayed them inside an interactive chart. Gemini offered choices to arrange reminders for what seemed to be an important ones, akin to RSVPs and a invoice, whereas providing buttons to archive emails it recognized as promotions. I requested Google Gemini to schedule a reminder to pay my invoice, and the AI assistant put it inside Google Duties with the proper due date. Once I requested it to pay the invoice, it navigated the billing interface and got here near asking me to enter my cost particulars, however (given the security concerns round agentic AI) I finished in need of letting it proceed.
Whilst you might simply set up your inbox manually, I discovered Gemini 3’s help considerably useful, because it dug up a couple of forgotten emails that I would’ve missed. It’s also possible to ask Gemini to search out and unsubscribe from spammy e mail suppliers in bulk, which is good.
Between Perplexity’s AI assistant, ChatGPT, and Gemini, Google’s AI chatbot (predictably) affords the richest integration with Gmail. Perplexity will pull up emails listed in your inbox, however you’ll want to inform it which of them to maintain, archive, or delete, as an alternative of simply hitting a button like you’ll be able to with Gemini. For some motive, ChatGPT refused to arrange my inbox, claiming its integration with Gmail is in a “read-only” mode regardless of readily sending an e mail by means of the app on my behalf. However whereas Gemini is straight related to Gmail, it was nonetheless far slower at sending emails within the app when in comparison with Perplexity.
Gemini nearly managed to ebook a restaurant reservation with out intervention, solely to falsely inform me that there’s a “price” related to making the reserving proper earlier than finalizing it. Once I requested concerning the cost, Gemini 3 backpedaled and stated it “seemingly referred to” the restaurant’s 16 p.c service cost. It then proceeded to ask me to verify my reservation thrice after which advised me there was a monetary transaction concerned once more. Sigh. Once more, I felt that I could complete these tasks far faster myself.
Regardless of the hiccups with finishing duties, Gemini 3 Professional’s interactive visualization options had been spectacular, and I might see how interactive fashions or visible layouts might be helpful in some eventualities — although I can’t see myself utilizing them every day, and Gemini’s text-based solutions are normally informative sufficient for me. For now, I feel I’ll simply preserve utilizing Gemini like I all the time do: for questions I won’t instantly discover by looking the online.
