with Gemini free. I mean, this is obviously the biggest story and that point is not even hard to argue because, you know, we don't usually look at benchmarks. And the reason for that is because they're not interesting and they don't push what's possible a lot. Well, that's a completely different story for Gemini 3. If we look at this, all of these black numbers that you see in this column, well, that means this is state-of-the-art on that benchmark. In other words, this new model that is available today inside of Gemini, which if you're not familiar is like Google's chat GPT competitor, and it's just bestin-class at everything. And I would like to point out these two, humanity's last exam, Arc AGI2. These two are generally speaking considered the hardest benchmarks on some of these. A year ago, models couldn't even get 10%. Now, Gemini 3 Pro without any tools got 37. 5% on humanity's last exam. And at this point, I would love to compare it to the other releases this week. GBT 5. 1 Pro and Grock 4. 1, the model from XAI, but they didn't release these numbers. And that's sort of how it goes. If one of the model makers makes a new model and it's not state-of-the-art on these benchmarks, they don't release it. But if it looks like this, oh yes, they will very proudly show you all the benchmarks on which they're crushing the competition. But again, this is just unbelievable. I mean, look at that across the board. The video recognition benchmark crushed general knowledge, agentic tool use. And one more note about the comparison of these and the benchmarks. Well, there's also LaMarina, as you might know, where people blindly test the models against each other and then rate them. And those leaderboards also show a strong preference for Gemini free pro. Look at that. On text generation, first place, webdev, first place. Vision first place. But for search, Grock 4 fast search is actually first. I think Gemini 3 Pro isn't even competing here. Overall, Gemini 3 Pro is the winner across all of these. Honorable mention to Grock 4. 1 and 4. 1 thinking in text generation. And then I don't see GPT 5. 1 Pro and the API variations on here yet. So I guess we'll have to wait a few days. So rather than looking further at that, let's have a look at what this thing can actually do. And generally I want to highlight what the community found this to be most useful at. It's obviously super released. This is just the first look at the model. But the general sentiment is this. It's a state-of-the-art model in most categories except coding. But the one thing that it went most viral on social media across the past few days for is frontend website design. So that's exactly what I want to try out here in our little test. All I'm going to do is just pull open gemini. google. com. You can see down here the new models are active already. They're the default. They don't even give you access to other models in here by default. And I'll just go to thinking with free pro and I'll ask for a website. So let's do something simple here. just create a stunning front end for design studio specializing in architecture that can be previewed in Gemini canvas feature that hosts the code and that will allow us to see the results right here inside of Gemini. While it does that, I'll do my default writing check. I want to see how this sounds. So, I emailed to my boss about the broken coffee machine. You might know this if you've been following the channel for a while. I just want to see its default email to get a bit of a feeling for it. So, I like how it opens the canvas here on the right side already. This word-like interface gives an explanation. All right. So, it's using variables as all the other state-of-the-art models. Now, it's nice and concise, and it actually provides solutions. I actually like this. This is super concise. A lot of other models are quite verbose here. This is all you need. All right, so that's a win. Now, let's have a look back at the front end. All right, let's preview this. Ooh, wow. Okay, so first up, notice how it also has images. All the other models would just create placeholders. Look at that. Isn't this a beautiful portfolio website? oneshot it from a oneline prompt and all the buttons work too. I mean this is just a first look but come on this is pretty damn good. No few more examples from across the internet with that you can get if you spend a bit more time. This guy built a Jarvis like HD with face tracking with object tracking and this guy built a 3D simulation of the Golden Gate Bridge in San Francisco. Or did he? Well, he did but he wasn't using Gemini 3. He was actually using the GPT 5. 1 Codeex Max at extra high setting. This is similar to GPT 5. 1 Pro that is now available in chat GPT. And also this model is ultra capable. I mean what we looked at a second ago here is if you use the developer model through the API then you can set it to this max and extra high setting. But my point here is yes everything you're about to see is super impressive but there's multiple models that can do it. Now yes Gemini wins on the benchmark but as per usual only time will show which one of these really becomes the fan favorite. And at this point, it's just ridiculous how they're competing. My point is, I guess all of these are good. And while Gemini 3 is the big story, just keep in mind that as I show you examples, there's probably other models, including Grock 4. 1, that could get close to these results, maybe not quite there. But I guess what I really want to say is that these models are getting so good that it's kind of hard to pinpoint the exact differences. What is possible though is looking at some of these example projects that Google provides, these are super cool, and you can just edit them yourselves and make them your own. So, I'll put the link to this one and four other projects like this below that they provide. So, this one is supposed to bring anything to life. And again, you could remix these to your own liking just by talking to the chat here on the left. But let's just see how this performs. Okay, I'll add this image of me speaking at the AI Advantage Summit. Beautiful. It's analyzing it and then it should create a world from this. — A few moments later. — Wa! Wait, so this Gemini coded application turned me into a game. Okay, passionate rant. Don't let your dreams be dreams. dreams. — See, you can kind of Okay, I want to hydrate a little bit. Slurp. Smart insight. Tell a joke. Why did chicken cross the road? Okay. Passionate rant. All right. It's a live stream simulator. So, this builds custom games from images you give it. Wow. How great is that? You can just go try this. Okay. What happens if I lasted 39 seconds and got,300 views. This is crazy. And you could take this application, customize it to yourself. All of this has been built by Gemini. I mean, this is really amazing. And then consider the fact that most people are even saying that this might not be the best coding model. That's the one benchmark where it's not the best, but clearly it performs. But another thing
And the next one was something really unexpected. I mean, it's been rumored for a while that Google wants to release a new Nano Banana. And yeah, we got it. It's not called Nano Banana 2, it's called Nanobanana Pro. And the main thing that I have seen that distinguishes this is the ability to generate a lot of text. I'm talking infographics with great detail and small text that is precise even though you have a lot of text. And it does it reliably. So, I really want to test that right now. Before I run a test prompt, I'll just show you a few examples. This is one that kind of captures the moment right now. Fair enough. I absolutely love this recreation in toothpicks from Apple lamps here on X. So, going from this image to this image. Look at how good that is. But we've seen restyling before, right? You could imagine other models do this. I mean, posters like this. That's amazing. Look at that. To be clear, this stuff was doable before, but you could almost bet money on the fact that there would be a mistake on it every single time. And this is just right. I really want to see this infographics in action. So, I'm going to create a brand new chat in Gemini where we tested Gemini 3 just a second ago. I'm going to say create an image and you're going to see it switches to the imaging tool. So, another way to do this is go to tools, go to create images. And now my only question is how do I know if this is Nano Banana or Nano Banana Pro? I guess I don't. Okay, so I want to do something creative and fun and I'll make it up on the spot. How about some Encyclopedia entry of a newly discovered alien fruit from a planet that only knows two colors, red and blue. Okay, see what nano banana can do. Ah, so it came up with the description, but I forgot to say generate an image. I thought if I select the image tool, it will do it by itself. Now we're making an image of this fruit, the dic chromosphere, also known as the etheladia picolora etheladira picolora. Okay, I'll spare you the rest of the details. Let's just have a look at this. And here you see it. Nano banana pro. Okay, so this is not what I was looking for. I was looking for a lot of text. I want a image of a NCopedia page with a lot of text and great detail. I'm testing the text writing abilities. This is not going to be enough. Oh, that's more like it. Okay, wait. Let's download this. I want to look at this in full screen mode together with you. So, definitely use Nano Banana Pro and the 3. 7 megaby large image is ready to be previewed. All right, let's really zoom in here and read this. Did I chromosphere in is an act two? Ah, and this is the first obvious mistake. I mean, it's pretty good, but yeah, these words aren't perfect. I can't claim that. I mean, maybe it could be excused by the fact that I'm asking for extraterrestrial fruit and it's making up words here. Let's do one more. And look, I'm not trying to say this is not good at text. I just want to see how good a text this is because that's the big claim. All right, full screen this. Zoom in a little bit. And here you can see a lot of this is just mumbo jumbo. I mean, to be fair, all of these headings, botanical description, cultivation, botanical description are perfectly fine. Even this one. Yeah, all of the ones that kind of matter. All the headings are great. Also, this is great. This is fantastic. I guess here there's a heading that's not quite right. And then all of these details are off. Obviously, figure 17 looks good. Images look good. So, definitely a step up. But going as far as saying what I opened the segment with that the text is perfect now in all cases, it's not true. You can rely upon it way more. It's clearly more consistent, but generating images of books is probably one of the harder things that you can throw at this. But I don't want to take away from how impressive this model is because it can do so much more than what I just showed you. Now, for example, in the release post, they showed this where you can upload a total of what is this? 15 different character images and then Nano Banana edits it into one image like this. I mean, this is insane. Or the character consistency where you go from one to multiple. I mean, to be fair, it was really good at that already, but now it's even better. I suppose the resolution is higher and you can see the ELOS coins in terms of image editing, it's just the highest. So, right, maybe the hype I brought in came from all over social media because they even say that, hey, can still struggle with small faces, accurate spelling, and fine details in images. But overall, super impressive. And let's be real, stuff like this was not thinkable a year ago and not possible a week ago. So, yeah, go try it out. See how you like it for
