NEW Google Gemini 2.5 Flash Update 🤯

Today we have the new update from Google Gemini 2. 5 Flash. So this is a brand new update. It's currently in preview mode like you can see right here and you can get access to it via the Gemini API in Google AI Studio or Vertex AI as well. We'll be comparing the benchmarks in a second. Bear in mind 2. 5 Flash is a thinking model, but you can actually switch off inside Gemini directly here.

So if we go to astudio. google. com, google. com. You can start using it. From what I've seen, it is actually free to use for a million tokens. So, you can see here it's got a token count. It doesn't seem to charge me anything for using this. And I think I've even got my credit card plugged in here. And then we can switch between thinking mode or non-thinking mode. And it's pretty easy and simple to use. So, they've basically said building on the popular foundation of 2. Flash, this new version delivers a major upgrade in reasoning capabilities whilst prioritizes speed and cost. So, let's grab some prompts from the AI profit boardroom link in the comments description if you want to get these and then let's start testing out. So, what we're going to say here is can you create a simple runner game in 3JS? So, we'll test it out right now. I'm going to leave thinking on mode on now cuz this is for like a coding task which takes a bit of thinking and reasoning to start using it. And it's now running its magic like so. Let's see how that performs. we'll come back to in a second in terms of how it's

performing. So it says Gemini 2. 5 models are thinking models etc on complex tasks that require multiple steps of reasoning. The thinking process allows the model to arrive at more accurate and comprehensive answers. In fact Gemini 2. 5 flash performs strongly on hard prompts on LM Arena second only to 2. 5 Pro. Now one of the biggest differentiators about this model is that actually it's a lot cheaper if you are coding with it directly. So if you're using the API directly from something like open router is way cheaper. All right. So for example here if we go to the models we can see Google Gemini 2. 5 pro previews available on open router. If we click on this you can see how much it costs and it is pretty cheap. So this is per million tokens by the way which is very cheap. We got 2. 5 flash preview there as well. That is even cheaper actually. This is the one we want to use. So we go to 2. 5 flash preview. You can see here 0. 15 per million input tokens output tokens 0. 6 and that was created on April the 17th. Now if we actually compare it and see where it's available to use. So we can use inside root code client all the usual suspects right here. And let's go back to Gemini inside. This model is

still coding out. We've got the code ready to go. Let's preview this on p5 js and we'll see how it performs. And actually this is on HTML. So we can just run the code right here. So run it inside LiveWeave. Test out. We got the car ready to go. Seems to be sideways, but let's see what we got here. There we go. It's coded out perfectly to be fair to it. The car itself is not great, but we've created a 3D game in like literally 2 seconds messing around with it. And one single prompt, one shot. Let's see if we can go off road here. Look at that. We're flying. absolutely flying. Not bad at all. It looks like we're coming to the end of the road here, but yeah, works perfectly. Nice. Cool. So, let's keep going through. You can see the comparisons of the models. So, we've got Gemini 2. 5 Flash versus 2. 0 Flash and OpenAI Claw 3. 7 Sonic Gro 3 and Deep Seek R1. So, if we compare the prices, as you can see, Claude 3. 7 Sonic is $3 per million tokens. Gemini 2. 5 Flash is 0. 15 per million tokens. So price and the cost difference is absolutely insane. It's performing pretty well on humanity's last exam. GP QA Diamond performing pretty well. So he's comparable. I always take these benchmarks with a little bit of pinch saw, but it seems pretty good at coding from what I've seen so far as well. Basically, this is the most costefficient thinking model, right? So you can code with thinking models but it is a lot cheaper and you can see how it performs right here. This is interesting as well. So they actually have a thinking budget that offers fine grained control over the maximum number of tokens. Right? So what does that mean? That means basically you can fine-tune this and set a budget for how much thinking you want it to do. Right? How advanced you want the thinking to be. That offers fine grain control over the maximum number of tokens. So the higher the budget, the more it's going to be. So, for example, you might have prompts that require low reasoning, medium reasoning, or high reasoning, right, when you're coding this out. And so, if something just requires a little bit of reasoning, then you can lower the budget and spend less tokens. It saves you money as well. So, it seems like you can get it inside Google AI Studio, which we were coding with before over here. You can also get it inside Vertex AI, and also it's available on Open Router as well. So, let's see how this performs. We're going to plug this into root code over here. So I'm going to grab this prompt from the AR profit boardroom. Let's open up recode. Grab that inside Visual Studio Code. I've got SPS on this if you want to learn how to set this stuff up inside the AR profit boardroom. And then from here we are going to go to open router and it's flash that you want to be using. Okay. And yeah, if you're wondering this model did just come out today, literally a few hours ago. So it's a brand new model. So we use flash preview. You can switch between thinking and flash preview here. So we'll select flash preview for now. Hit save. Hit done and then inside here we'll start coding out. So just going to use this

prompt to generate a landing page and see how that performs. Should give us a price as well as we go through or we can check the token count on open router if that doesn't give us a price. Pretty fast. So you can see the cost here. It's just ridiculous. It's not even a scent. It's cost us so far 0. 0083 to run this and we've just built the website. Seems really fast as well. Let's open that up. See how we get on. So that is the website ready to go. Obviously, it requires a lot more back and forth than that, but just to build a simple website out, pretty fast, pretty easy to do, super cheap. Wouldn't say it's as good as Claw 3. 7 Sonet, honestly, when you're building stuff out with that, but it does the job. Let's have a look now. We'll test this out for building out a calculator landing page. So, let's grab this. In fact, first thing I'm going to do is just say, can you improve the UI? Make it more colorful, sleek, and modern. I'm impressed with the cost and the speed. I'm not impressed with the UI so far. I still think Gemini Pro is going to do something a lot better than that. Let's go back to Gemini as well. So, we can use this inside Gemini directly. So, if you go inside Gemini Advance, which is a paid model, you can switch between 2. 0 Flash, Flash Pro is the one that just came out today. So, it's 2. 5 Flash. And then we can also use a canvas here. And I think that'll give you better results for tools and that sort of thing. Right. So, if we grab this and we plug that through Gemini, honestly, I think you're going to get much better results. But let's see what we get back here. And the good thing is if it's a simple HTML canvas sort of thing, then you can quickly just build it and preview it inside the canvas here. So, once this is done, we can click on preview and then test it out. Some people are saying this has comparatively low benchmarks. It depends what you're comparing it to, right? So, for example, OpenAI 04 Mini is comparatively lower, right? But if you actually compare it to Claude 3. 7 Sonic, it is holding its own right. So for example, humanity's last exam, you can see here 12. 1% for 2. 5 flash. It is outperforming Claude 3. 7 for GPQA diamond. It's pretty much the same as 3. 7. Let's have a look what else we got here. For maths, it's outperforming Claude by a long way. I don't think you're going to beat open Mini, but look at the price difference. It's literally almost 10x, isn't it? So, you know, that's the difference right here. When I've tested out 04 mini as well, it takes ages to code with it and it just wasn't that impressive. You check out my other videos on that. So, we've got this back. Let's see what we got here. That is looking a little bit more like it. Honestly, it's pretty average. Let's have a look inside Gemini Advancer and see what we got here. Oh, that is just even worse, mate. Even worse, Sunshine. All right, let's test something else out now. So I'm what I'm going to see is does this give us better outputs if we just ask for a very basic prompt and we don't make it too complicated. So if I say okay create the best possible SEO calculator you can with a beautiful plus changing color background and information selling my SEO agency on the page. Let's see what that does. It did pretty well in the game. It did pretty bad for you but let's see what it comes back with on this. So, I wonder if it's good for creating content as well, cuz it's outperforming Claw 3. 7 Sonic, which is one of my best and favorite tools for creating content. So, we'll test that in a second. Yeah, that's just particularly average, isn't it? I think for UI, you are you going to use that? It's definitely not It's not the best. It's not the best, mate. I'll tell you that for free. If we switch over to Let me compare this side by side versus Pro. Right. So, if we go over to Pro right here, and then we use exactly the same prompt. I can almost guarantee you're going to get better results, but let's see what we get back. So, this is Pro Experimental and we're testing the same prompt, but side by side versus Flash. I'm going to ask as well in the meantime, just make the U much nicer and more modern and sleek cuz you're never going to publish that website, right? That's even Bolt or something could do better than that. Some people saying, I think 2. 0 Flash is better than that. I would agree with you there. I think genuinely I've not seen bad that bad UI for a long time. Top crowd. I just honestly like even on the review I it's not a great UI. It might be really good for other stuff like for example coding the back end or for example creating content which we'll test in a second but for coding out UI you just wouldn't publish that, right? Like it's way too basic. That gives you a much better style. You can improve the color contrast in right there. So if we say to it now, okay, just make the colors more contrasting, eg the font versus the background should be contrasting and easy to read, make the font more modern, but that definitely looks better than 2. 5 Flash, which after a couple of revisions looks pretty bad. So I got to say I'm disappointed on the UI side for sure. So, what I'm going to say

else out now. So, what I'm going to say is make me a captivating endless runner game in p5 js. We'll do that inside AI Studio while the other two are coding out. Good thing with Google is that you can code in multiple tabs and then whilst I'm talking to you, you can see the outputs from each. So, we'll see what that comes back with as well. All right, so that was pretty trash. That's the output from 2. 5 Pro, which I'd say is definitely better. Let's have a look what we got back from AI Studio here. In the meantime, we'll run a content prompt as well. So, I'm going to grab this prompt right here, and we'll see how it performs for creating content. So, we'll run Flash Experimental and test it out. Again, though, just to reiterate, cuz I know some people will say, "Oh, this is not free, blah, blah. " I'm using 2. 5 Flash preview for free inside AI Studio. Gemini Advanced is a paid model, so you're not going to get it free there. If you use the API, you're not going to get it free. But if you code inside AI Studio directly using Flash preview over here, then it shouldn't cost you any money from what I know. So I don't even have my billing details in here. So we got the P. Why is it coded in separate files? Here we go. All right, let's copy this. So we got to P5GS and just test this. Plug that in. Hit play. Dino Runner Pixel Dino Runner ready to go. Let's see what we get. Background's not bad. Game is working perfectly. UI is not great, but that's a oneshot prompt, which is pretty impressive to be fair. I don't like the colors. Again, I think it does struggle on UI stuff. Like, I think that's what it's probably the worst at so far. Let's see how it does on creating the content. So, let's talk about Gro A and what it means to anyone serious about SEO and creating banging content. You've heard the noise, right? Another AI blah blah. Yeah, the content is quite humanized. It reads decently. It's frontloaded the keyword. It's a pretty long article as well. Let's see how long that is with one prompt. Whoa, it just blasted out 2,000 words in one shot. That is crazy. Yeah, for content writing, it's pretty good as well. Let's just test it out in terms of AI detectability. Just going to plug this in here. I know, for example, when I've tested this out with other models recently, they come out undetectable completely. That one is not bad. It still passes the AI detector, but it's coming out at 12. 3%. Which is not as good as I've seen from other models. Let's see how it's performing on LM Marina as well. So, if we have a look at the leaderboard here, we've got Gemini 2. 5 Pro again. Yeah, I would agree. I think 2. 5 Pro is definitely a superior model. Obviously, it's more expensive to code with. Chat GPT coming out second. Gro 3 and GPT 4. 5. And then you got Flash Preview coming in at what joint second right there. Not bad at all. Google are dominating. Look at the top five. Google and OpenAI absolutely dominating. Let's just go back to the benchmarks to see which one is the best for maths. So, we got the Amy maths test over here. So, if we have a look, Gemini 2. 5 Flash 78% Gemini 2. 0 Flash, there's a huge difference. It's a massive step up from its previous counterpart. OpenAI04 mini though is just destroying maths. Look at that. Like, it's absolutely destroying everyone. Gro 3 beta says 93%. But when I've tested Gro 3 recently, I wasn't that impressed to be honest. So I think if you're going to pick any model for maths, just go with open AI04 mini. Seems to be a lot better. Let's see how much it costs in our end. If we go back into open router and we'll go inside activity and just check what the pricing is. Well, when we're using it, to give you an example, right, if we're using Claw 3. 7 Sonic, it's going to rinse us for credits. So, we start using something like Gemini 2. 5 Pl. So, it's super cheap, but also look at the speed here. So, you see the speed, it's 142 tokens per second. Compare that to something like Claude, which is 53 tokens per second when I've used it previously. GPT 4. 1 82. 4 four tokens per second. So, you know, in terms of speed, I don't think you're going to beat Flash from what I've seen. 199 tokens per second there. So, it seems a lot faster and it's a lot cheaper to code with as well. I want to see as well, is it directly available inside NA10? So, we'll go inside our workflows inside NA10 and just have a cheeky gander here. Sign in, see what we got. So, let's see here. I'm going to clone this Optimus Alpha AI agent. We'll duplicate that and we'll call this Gemini 2. 5 flash agent inside NA10. If you want to get all my best workflows, by the way, for NA10, we've got tons inside the AI profit boardroom. Just go inside the classroom and then go to the NA10 template section here. And you can see like we've got a redonulous amount of NA10 templates that you can steal from us. So, let's have a look here. We're going to switch this now to Google 2. 5 flash. And we just use preview not thinking and just test this out. All right. So this just super basic agent just like a helpful assistant. I just want to see how it performs, how fast it is, etc. Yeah, it's pretty quick, isn't it? It's pretty quick. That might be one of your best AI workflows for creating like customer chat bots and stuff on the website cuz you could just deploy this. It's going to be really fast. You load up the AI agent with all your best knowledge, etc. inside the memory or inside the system message. And also, yeah, you should be good to go on that. Really easy and simple to set up. Let's see what other people are saying about this. So, if we go over to Twitter, so Logan announced Gemini 2. 5 flash here. Our first unified reasoning model with thinking budgets. I do the fact that it gives you options on thinking budgets and that sort of thing. If you compare here, so this is the arena score. So, LM arena score versus the price. And you can see 2. 5 flash is right at the top in terms of arena score, but also way cheaper than your average sort of models to use. Let's give it something a bit more advanced now. So, what I'm going to say, and this is something that works really well inside Gemini 2. 5 Pro. If we go to the SAP section, then we're going to scroll down and to the master list of prompts and we're going to create an interactive water molecule simulation. So to do this, I'm going to use canvas inside 2. 5 Flash. See what we get back. In the meantime, just in case that doesn't work, would you use the chat? Also, what's interesting inside AI Studio now, they've separated each model. So you got 2. 5, Gemini 2 preview, Gemini 1. 5, and Gemini Gemma. Right. So if we go to preview models, you've got a couple of options there, but obviously we want to be using 2. 50 preview. Now, for this, we're going to use the thinking mode. We can set a thinking budget so we can switch this around here. And then let's plug that in. Yeah. Enter. And see what we get back in a sec. There we go. That's pretty good to be fair. Let's speed this bad boy up. Slow it down. Yeah, that worked perfectly, didn't it? Pretty good. Let's build Flappy Bird. A classic. By the way, if you want to see like real business use cases, because obviously these prompts are f fun and nice and that sort of thing, but if you actually want to see, okay, how we implementing all this stuff inside our business and actually making money with it, then check out the crash course inside the classroom with the AI profit boarding. So, if you go in here, you'll see all these different methods that we've used to automate. So, for example, like automating emails, our customer service and sales team, they've been able to automate a lot of their emails using this module right here. We have Twitter automations, newsletter, shorts, AI avatar videos, Fathom, Zappier, etc. Like loads of cool stuff in there if you want to check it out. So, let's go back. We've got Flappy Bird ready to go. Is that working? It doesn't seem to have done anything, right? Yeah. It's okay. It's not amazing, but I think it'll be really good for developers because if it's cheap to build apps with, then obviously SAS is going to become a lot more profitable to build AI apps with, and so that could lead to good things. But from what I've seen, it's okay. It's not as good as Gemini Pro 2. 5 Experimental. It is cheap for developers. You do seem to be able to use it for free inside AI Studio. It can give decent outputs. Like for example, the interactive water molecule simulation, but would I switch between these? I probably wouldn't use it in videos. It's probably just going to let me down. When I'm using it inside automations, it's going to let me down quickly. but it might be really good as well for NA10 workflows because it's super fast and cheap to use inside there. All right, so thanks so much for

watching. If you want to get all my automations and practical use cases of how to grow your business and make more money and save time with AI, feel free to get the AI profit boardroom link in the comments description. Comes with all my best automations for email, social media. All this stuff I'm actually using inside my business as well by the way automation and 10 templates, my crash course based on everything we use. There's also weekly Q& A calls here. So, if you ever get stuck or lost or whatever, you can jump on the Q& A and ask any questions you have. And inside the community, if you ever get stuck, you can ask 673 people for any tips that you have. And you can see here, we're just constantly updating this. So, we've got tons of updates that just came out this week if you want to check that out. Additionally, if you want to get a free onetoone SEO strategy session, feel free to get that link inside the comments description. and we'll show you how we take websites from zero to 145,000 visits a month and generate hundreds of thousands of dollars on autopilot with SEO. Feel free to get that. We'll give you a custom tailored game plan and you'll learn exactly what's working for us based on all our happy clients like you can see right here. Thanks so much for watching. Appreciate it and I'll see you on the next one. Cheers. Saba.