Gemini 3 Just DROPPED & It's INSANE!

I basically think Gemini have won. Long last, Google have rejoined the chat with their latest bombshell, Gemini 3, and it's exploded onto the scene with a plethora of products, including the slightly disastrous, I think, anti-gravity code editor. So, this is something we've been waiting for for a long, long time. 2.5 was released about five, six months ago, which is crazy when you think about it because at the end of the day, you know, we're used to waiting kind of multi-year for like new drops, but this is the age of AI. This is where stuff moves fast and it's completely multi multimodal. So, video, images, text, um, as as inputs, which is incredible. It's killing every single benchmark minus one kin Kane which is the sweet benchmark and this is the coding benchmark. The vending bench which is a really interesting one. This is where the AI is tasked with just managing a vending machine and it's got to manage profits and loss and you know pricing and all this kind of stuff. I'd love to see it in action. Sonet was the leader of that earning three nearly four grand whereas this one is pushing 5 1/2 grand. and it's earned from the vending bench which is really cool. This is bridging on scary humanity's last exam which is a really really it's supposed to be the most difficult test for an AI to do. It's got to at real time kind of figure out a pattern which a human can do. 37.5 without tools pushing 41% with tools. Now the current leader was GPT5 Pro uh at 30 30.7. So huge huge jump. I mean look at Sonnet and Gemini 2. Look how far behind Gemini 2.5 Pro. So this is this is actually insane that we're getting to this sort of level. But it's uh yeah it's crushing in pretty much all of the benchmarks. And I wanted to show you what it's like with its web design capabilities because they gave my favorite code tool replet early access. And I gave it the task of designing a website by comparison. The I think you need to look at the details or not even the details but just you know the the collection of everything before like whereas before it would design things and it would just kind of just it basically be a glorified wireframe. There was no creativity. there was no nuance to like what the app you were building and the design of itself. Whereas I feel like this is very in tune with the recipe app. Little things like the button here as kind of like this dotted stuff. It feels very paper, very textured and things like that. There's a correlation between what I asked it to do, what I asked it to design, and the actual output. This is using a new design mode inside of Replet which apparently is completely free at the moment and you can get it just to design your thing and this is this is kind of what I did. They've also chucked it into Stitch. Now I wasn't that impressed with this outcome. So here's my website. The thing I told you that I've just completely vied yesterday before uh Gemini 3/ Nanobanana 2 was put into this. I asked it to redesign this website and first of all I said there's a YOLO thing when you ask it to redesign this YOLO and this is what it came up with. This is not Nano Banana 2. This is just the normal uh this was whatever it was using before maybe 2.5 Pro or 2.5 Flash kind of like this one. This is kind of cool. Um yeah, but yeah uh this is kind of the design. And then I I wasn't too happy with that and I was like, well, just just do a slight redesign on what's already there because you can again choose these different things and not too much difference there. Just a subtle things, different change of color there, different change of fonts there. And then this one completely broke. So the good thing with Stitch is that you can actually, and this is what I did, you can download the HTML that creates this. This is this is fully responsive. Like if I just Oh, that's cool. Can you I lost it. So there I mean it's super long but like you know if I do this thing it starts to become more sense thing you know [snorts] um and then I reran it earlier today with Nano Banana 2 and I wasn't too impressed. I mean this is kind of what I came up with. Not really. Yeah. Certainly more creative I think. I mean look at that. It's given me a satellite thing with like you know different areas and whatever. Uh this is like HTML like with CSS everything is ready. Yeah. Yeah. Exactly. So I can actually um export or view. Well, this is actually very interesting. I can put it into Jules. I can put it into AI Studio. So here's the HTML that makes up the design. They've kind of pushed and pulled like this Figma integration. But this is all the code. You can copy that. You can download it. You can do whatever you want in it. So um yeah, it's pretty it's I like this tool. It's pretty cool. And then yeah down here it's like a redesign thing. So what they've added is that you can I can select this redesign with nanoban 2 and then actually provide some sort of design some sort of prompt that actually what I wanted to do. So a little bit more fine grain control. So have you got AI studio up there then this project also? Oh yeah. Physics simulate. Did you prompt this or is this just a template? No, this is just from the what's it called? like they have like some ideas that you can remix and they are okayish. Some of them are fun. So you saw this one on the Google article already mentioned and it's a game apparently. Oh yeah, I saw something like this. So they actually did a live stream when they launched Gemini 3 and I was watching it. It was AI Studio and they just got a bunch of people just to create a whole heap of stuff and there's some incredible stuff as you're seeing here. It's good at like oneshotting things and and just creating stuff. But I think once you get into heavyduty kind of like real world coding with linking to databases and and and building real life applications, this is where I feel like Gemini falls short. It's just not It's just not Yeah, this was one of the games they uh they showed. Yeah, it's a it's a camera. You need your camera. You do this in front of the camera with your hands. This is that Oh, yeah. So, you swipe down, swipe left, and swipe right. Look. See? You see a bit Beat Sabers? Yeah. The the the latency sucks. To get this sort of game to, you know, with natural language is kind of crazy. Like use my camera to track my hands and like this is this is wild. But I do think that Gemini 3 is a a big a big deal much bigger than uh GPT5 much bigger than GPT 5.1 because that was just a tweak obviously because it's a minor number um for for normies for like AI just in general I think this is such a significant leap forward that I think it's going to be really hard to beat in any significant way. I think we'll have minor adjustments here or there. I can't I don't know. I will probably eat my words, but I just cannot imagine AI getting much better than Gemini at what Gemini 3 is right now. I think we'll get marginal increases. I'll be very very surprised. I basically think Gemini have won. People seem to be very uh surprised by the quality or like the creativity and the looks of the code. that um Gemini 3 is outputting and especially with anti-gravity people have been using it but I I mean it doesn't matter it doesn't have to be with anti-gravity but yeah I saw a bunch of stuff people shared on Twitter but again a lot of people were also sharing like obviously like creative and custom work and they were like look what I did with one single prompt like just being cheeky and no disrespect to these people but who are Are they like are they hardcore developers? No, no, no. People were like joking about it. But the quality [clears throat] what I think why people were interested in the whole conversation was because for some reason when when we asked Gemini to output whatever type of code when it's interactive and 3D, it's just a bit more creative. And again, I think this is why Replet created this design mode because it's um I mean, look at this. You know, it's no longer wireframey. And again, you're not going to you know, I'm not a designer, but this is totally 100% good for me. Like I if you want to design, you're you're going to you're going to you know, you're going to do your own thing and you're not going to want to vibe design anything because you want a very specific thing. But for someone who just wants something that looks kind of cool, uh, this is absolutely great. And that was the first thing I noticed when I was checking out Gemini 3 is the creativity aspect. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter.