Released: Google Launched the New Gemini... But Where's the 3.5 Pro?

[00:02] not one, but three new models—but it's already generating a lot of buzz within the artificial intelligence community. That's because everything is moving very quickly. So, if you don't want to be out of date, watch this video until the

[00:14] will be very useful for you and you'll understand if these models that Google launched are models to compete with the top-of-the-line models on the market today, or if they are simply new models launched to meet a

[00:28] content. Well, three models have been released. The Gemini 3.6 Flash, the been released. The Gemini 3.6 Flash, the Gemini 3.5 Flash Lite and the Gemini 3.5 Flash Cyber. Each of these models serves a different purpose. You

[00:43] can now get immediate access to these two models, the 3.5 Flash and the 3.5 Flash Lite; they are already available. If you access your GMI there, you can already select those models. But what matters is what they

[00:56] are able to do. This is going to surprise you a lot, okay? So watch not yet available is the 3.5 Flash Cyber. This is a model that is being monitored, and it's more focused on finding security flaws,

[01:10] Google promises they are smarter and faster, but are they some tests here. We're not going to just look at Google's benchmarks here, because what really matters are the tests done in practice. But if you want,

[01:23] you can go there later. Here it talks a little more about the speed of the models, it talks a little more about the comparison. Google claims that version 3.6 consumes fewer tokens than version 3.5. I've been using version 3.5 in some projects and it

[01:37] look at our tests and comparisons. To start things off, since we 're still in the World Cup finals mood and all Brazilians are very happy about it, I asked Gemini to create an SVG animation that you

[01:52] saw here until recently. The models weren't capable of creating these animations first, and then I'll show you which model made which animation, and you'll be surprised. So here we have our first animation that was

[02:05] made, showing the scoreboard, right? 1-0 for Spain. I think every Brazilian was really rooting for Spain to win, right? Then you click on download. You can download this animated SVG here. And you can

[02:17] even make some changes, and there are even some sound effects you can how it works. Like a Spanish bull, that's what I asked for. And here's an Argentinian condor over there, sad because it

[02:29] lost. Here we have the same animation made by a different model. Look bull is pretty crazy, isn't it? He even put down the World Cup trophy. And here we made by a different model. Which model do you think is the 3.1? Which model do you

[02:44] think is the 3.5 and the 3.6? This will surprise you quite a bit. Google continues to claim that version 3.1 is their top-of-the-line model. But look, this animation here was made by version 3.1 and

[02:57] it took 182 seconds to generate. You can see that the 3.1 is really far behind the current models here. This makes me think that maybe Google is working very hard on Gemini 4, which is what people are

[03:11] reports that they are working on it, but some people are saying that version 3.5 Pro is coming out. That doesn't make much sense because they've already released 3.6, so I don't know. Here we have the same

[03:23] animation made by model 3.5. Look, the explanation is even here, right? The Condor and the Champion Bull. And obviously here we have the animation done by 3.6. It's a little better. In my opinion, it's

[03:37] not that different, but it's a little better. This one took 299 seconds. This one took just a little less time, 252 seconds to even the animation; what surprised me, I'm about to show you, was something I'm about to show you. But hey,

[03:51] go ahead and comment below where you're watching this video from so I can send you a hug. Whether you were happy or not with find out. Let's go. Look, I asked these three models, always using the

[04:04] that would represent what it's like to play Mario. And I asked to be able to give instructions at the beginning on how the game worked and everything else. And here, look, have two phases. That's what I asked for, two phases. So, we're going to try it

[04:18] here, phase one, we have it here, there's even a little noise there, right? Very similar indeed . Make the little dots, see? That's great, man. Imagine, you can build this on your computer with just one

[04:33] command prompt, right? [music] Look how cool that is, right? Let's see. I lost my life there. So, this one here , in a little while you'll see which model it was made. Start counting the points up there.

[04:47] Ready. It looks a lot like those games from the 90s, doesn't it? Now I comment here which one you thought was best. That just got interesting. He talks about playing, testing the level, and jumping. Let's play. Oh, I started dying right there.

[05:00] It seems like the drawing looks better in this one , right? Let's see. Which model do you think made this one? The little turtle is right over there. Ready. Let's

[05:14] test the next one here now. Wow, what a difference. Look at this. There's no noise, you know what the result is. Let's start with this one. This one here, look, I'll click here so we can see it. It was exactly the same prompt, okay?

[05:28] prompt here. I'm creating a Mario game that I can play myself, using only HTML. I've already explained everything. JavaScript and everything else. This was the Gemini 3.5 Flash. It took 399 seconds to make this little game that I could publish, and

[05:44] you wanted, okay? I even liked the little drawing on this one. Oh, it's super easy to use and it even has that little beeping sound and everything. This version here was made using

[05:56] model 3.6, and it took 229 seconds. So you can see that it's you can see that it's actually faster, look. 339.229 seconds. I thought this one looked a bit more like the old Mario design, but the

[06:08] other one's design is better. Believe it or not, the gameplay in 3.5 is better than this one despite being faster, it doesn't seem to me to be that much better than the 3.5, I don't know. And obviously this one was made with Pro 3.1, it didn't take long,

[06:22] 172 seconds, but it turned out awful. think. I really didn't like this version. Obviously, here we have the 3.1 Pro, which is very good, and used to be very good. In UI, you know, to be able to create

[06:36] visual interfaces. Now it's becoming a thing of the past. And Google says that these new models agents. So, many of the agents you're using online today, across various platforms, utilize these Google templates to manage everything they

[06:51] need, because they're fast and reasonably affordable. Version 3.6 uses significantly fewer tokens and is also cheaper for everyday use at Gemai. This is more for those who really want to build with this

[07:04] model, but it's also a model that navigates the internet much better. And what 's great is that this model also has updated knowledge up to March of this for you to use. There is now a certain amount of rejection. If you look at the

[07:18] comments on the previous post, you'll see: "Hey, where's the new 3.5 Pro model? Will they release the 4? What's going to happen?" People are wondering why Google is actually lagging behind OpenAI or

[07:30] Anthropic? Apparently Google is already training the 4th generation, but hasn't even released the 3.5 Pro yet. Would releasing the 4th generation directly make more sense? I do I can know what you think. And will it be as revolutionary

[07:44] it be as revolutionary as GPT 5.6, or was Fable 5 from Antropic so powerful that it was even blocked by the government ? The bar for rising very rapidly, and the Chinese are pushing it even higher.

[07:58] videos with the newest, Chinese models that have been released. I'll post content here twice a week and also on my Instagram. If you don't follow me, go check out @rop.bver on Instagram, where I'm always posting

[08:12] you thought in the comments below. I didn't think this Google launch was anything fantastic. They are trying to show something so as not to fall behind. That's just my opinion, obviously, but I hope they really bring out a top-of-the-line model

[08:25] models. And what Google always does is bring a model that is also accessible. you're not subscribed, subscribe to the channel.