ChatGPT vs Claude vs Gemini: The Best AI for Each Use Case in 2025

Hey everyone. So today I want to share a super practical guide on the best AI models by use case. In this video I'm going to test chat GPT versus claude versus Gemini head-to-head for coding, writing, deep research, multimodal, and other use cases. As you can see in this table, all three models are good at different things. But don't take my word for it. Let's run through some quick

demos now. Okay, let's start coding. So, I asked each model to create a full feature Tetris with beautiful graphics and controls. And let's run through what each one made. So, let's start with Claude. So, here we have Claude and let's open the Tetris game. And you can see here that Claude built a beautiful feature complete game with scores, previews, and great controls. Let's make sure it all works. Let's try to finish off a line. Okay, I need to work on my Tetris skills. All right, looks great. So, that's Claude and it pretty much did this in one shot. Again, I just asked it to create a fullfeatured version of Tetris with beautiful graphics and controls. Now let's check out chatbt. Okay, so I use chat03 which is the best reasoning model that catch has. And let's open up canvas to see what this game looks like. I open canvas. You got to click on preview to see it in action. And you see here that the game works, but it's nowhere near as fullfeatured as the clawed version. So last but not least, let's check out Gemini. I gave the exact same prompt. use Gemini 2. 5 Pro. And here's the game that Gemini made. So, let's press start. And yeah, the game works great. It's not as polished visually as Claude is, although it's pretty close. So, I think right now based on this test, Claude is the best at code coding. And in this case, I want to use Claude Sonnet instead of Opus, which is even better. All right, because coding is fun, let's try another test, right? So this time I asked each model to create a 3D interactive solar system and let's see what each model made. Let's start with Claude again. Okay, so again the prompt is just to build a beautiful 3D solar system simulator, make it as realistic and interactive as possible. And you see here that Claude made a pretty awesome simulator. The planets are a little bit dark, but I can control the speed. scale. And what's even more, if I click on the sun or different planets, it tells me information about each sun or each planet. Right? So, this is a pretty awesome educational tool that I can explore with my kids to learn about the solar system. All right. Now, let's check out Chat GPT. So, unfortunately for Chat GPT, I couldn't get it to actually run. Even asking it to fix errors. Let's try opening up the canvas. and preview and unfortunately it doesn't work right. So in this case ChachiP didn't actually see to build the solar system. Now let's try Gemini. For Gemini it is succeed to build a solar system and some planets like Saturn look actually pretty great with the rings but like the planets are like way too small. I can barely see them. Maybe this is like a realistic view of what the solar system actually looks like. The planets are pretty spread out, but they're not really great for education. A little bit too small. And some of the planets don't have any kind of like info panel or anything. So again, in this case, I think Claude is the winner here. And let's go further, right? I really wanted to push to see what Claude, especially Claude Opus, was

capable of. So I asked it to build the first level of the original Super Mario game. Not sure if you're old enough to play the game on the original Nintendo system, but it's one of my favorite childhood memories. So, let's open this. And what I did was create level one of 2D sides scrolling Mario style platform game. Gave it some instructions. And I also included a image that I found on Google search with the map of level one of Super Mario. Right. So, it took a few back and forth. I had to ask it to make Mario and the glmbbas look better because in the beginning it was just a bunch of squares. I asked it to change the floor and give it some instructions like for example it made something where I can only see half of Mario instead of the full character. I also asked it to move the flag placement based on this image and even add a mushroom. Right. So now let's go ahead and play the game. And this back and forth probably took about 10 or 15 minutes to do. But what's incredible is, as you can see here, it's actually made a pretty incredible replica of level one of Super Mario. And I just ate the mushroom. I grew bigger in size. There are glmbbas going around. I can keep going here. I can jump on the glmbbas. And I can even break blocks. Although it breaks the whole block at once. I can keep going. And yeah, if my memory serves me correctly, this is a pretty accurate representation of the first level of Mario. And let's keep going. Let's keep going all the way to the flag. So, let's go. And here we have the flag. And there you go. Level complete. And I wasn't able to get the flag to animate just yet, but I'm sure with a few more prompts, I can get it to all work. And this is pretty damn impressive, right? Like I'm not even using cursor or any of those vibe coding tools. I'm just building this in cloud itself. So, the fact that Claude built an actual playable level one with mushrooms, glmbbas, and working physics is just incredible. So, if you have Claw 4, definitely try making something like this. Maybe make 2D Zelda or make some retro game like F-Zero. Have some fun with this. Okay, but let's go back to our slides here. And we've shown that Claude is the best coding model right now. But here's a catch. Claude is almost 20 times as expensive as Gemini 2. 5 Flash, right? So you see here that Claude costs $3 while Gemini 2. 5 Flash only cost 15. So if cost is a factor when you're building your AI product, I think personally that you can't go wrong with Gemini. It's not as good as Claude, but in many cases it's good enough. And furthermore, Gemini also has a larger context window than either Claude or ChatgPT. Okay, so I guess go for Claude if you want the best coding model and go for Gemini if you want the best price toerformance ratio for coding. All right. And I also want to show you some other things that Google has been cooking lately. I was like super impressed by the recent IO where they launched what seems 100 different AI features. And one of the most impressive ones that I got early access to is Gemini Dusion. And what's really impressive about this is it makes Vibe coding feel almost instant. So let me give you a quick demo right now. Okay, so Gemini Dusion basically lets you code lightning fast and create apps at lightning speed. So let's give it a quick run through. Right. So I'm going to ask it to create a brick breakout game. Make the graphics look awesome and make the brick break into particles when it's done. All right. So, you see here within a couple seconds, it's made the break breakout game and it seems to work just fine. And you're not going to be able to get this model to build Super Mario. But for simple apps and simple games, it's just a great feeling to build this stuff at incredible lightning speed and to iterate fast. Okay, so that's it for coding. So, TRDR choose Claw 4 for the best coding model and Gemini 2. 5 if cost is a factor. All right, let's move on to writing, which

is actually my most common AI use case. So, I have a prompt that I use to ask AI to edit my newsletter posts. And let me show you what this prompt is. So, it basically has a bunch of instructions up front and what's really important is it has examples of my best newsletter post that I wrote in the past. So that this way AI can understand my writing style and use that to edit a draft post. So here's a draft post that I created. It's basically like a huge wall of text. And another tip is that these days a lot of times I create the first draft of my writing by just using voice dictation because it's way faster to speak your mind than to stare at a blank page. But basically this is a draft post and then let's use my prompt to see what the output is for the three different models. Right? So let's start with Claude again. Okay. So what I did was I pasted in my prompt and first I asked Claude to analyze my writing style and here is a bunch of style summary and then I pasted my wall of text draft post and here's Claude's edit and you see it does a pretty great job of keeping to the style that I like. I always start my post with dear subscribers and today this is what we're going to cover and cloud did a good job of that. It added section headers. It kept to the kind of like short paragraph and concise style that I like to write in. So that's Claude and this is pretty great, right? And now let's check out chatb. So again I did this same thing. I gave it my prompt and then draft post. And chatd did a decent job. For example, for the introduction, it basically cut everything down into just two sentences, right? Two or three sentences. And like I almost never write introduction that's that short. It's definitely much longer than that. And yeah, I think overall the post just feels a little bit too short. Like it doesn't feel as meaty as my original draft is. Now, let's check out Gemini. Same process. And here's Gemini's post. And I think Gemini has the opposite problem of chatbt. The post is a little bit too long. I think it goes into a bunch of detail and some of the writing here is not necessarily close to my style. Like I don't really say, "Hi everyone, your name here or Peter here. " It's weird, right? So for me, Claude is still my daily workhorse because it consistently captures my writing style better than any other model. and writing and editing is by far my most common AI use case. Whether that's editing newsletter posts, editing PRDs, or just editing a short paragraph, right? So, for me at least, especially if you give Claude examples, it's the best at closely following your style. Okay. So far, we've been seeing a

lot of praises to Claude and Gemini. But let's give ChatGpt credit where credit is due. All three models are capable of answering everyday questions. But chat GBT has one feature that the others don't. Memory. And memory is really a magical experience. It leads to magic. Like chat GBT actually suggesting what I should ask about. For example, I'm about to go on a trip to France and Chat GPT actually suggested that I ask about the best places to visit in southern France. And it's also really great at helping you be introspective and find your hidden talents and blind spots. So try a prompt like this. Tell me something special or unique you notice about me, but I haven't realized about myself yet. Doesn't have to be something positive. nice. Just be truthful. Try something like this in chatbt, especially if you use chatb a lot. And the answer it can give you is going to be pretty amazing. And if you want to go super hardcore, try giving ChachiP this prompt to have it be a brutally honest advisor. So basically, it's like, I want you to act like my brutally honest advisor. Speak to me with someone with massive potential, but also a lot of blind spots. Just be super honest and blunt with me. And when I put this prompt to chat, it told me that I was bad at prioritization and trying to do too many things at once and limiting my potential in whatever I'm trying to do. and it might tell you something entirely different. I've had people who have messaged me telling me how this prompt was a wakeup call for them. I've included a link in the video description to a post that has the full prompt here that you can copy and paste. So, definitely check that out and see if chatbt can give you some rude awakening or brutally honest advice for yourself. All right, now let's move on to deep research. So every provider now offers

deep research features. So I tested each model with this prompt. Research the AI coding market and share Bolt's strategy to compete. Bolt is an AI prototyping tool that I really like using. Let's open Claude first. I think Claude's deep research is still only available to people who pay $100 plus for the premium plan. But basically, this is my prompt. Come up with a strategy for to compete in AI coding space. Here are some things to look at and let's take a look. So, it spent seven minutes researching 427 sources. That I think is by far the most sources that any deep research looked at. And let's take a look at the report. All right. So, what's really interesting about Claude is that even though it looked at 427 sources, the report is surprisingly short. Maybe it's seven pages or so, right? But let's take a read through. It's saying that boat's unique combination of accessible interface, powerful tech, and early market position provides a foundation for dominance success requires evolving beyond build apps in minutes to become a trusted platform for AI augmented development. So that is pretty vague. I don't really know what that means, right? What is the actual strategy that Claude is suggesting? Okay, so have technical superiority through web containers. Build trust through transparency education. create network effects through ecosystem development and enterprise features. I think of the three suggestions, the last one is probably something that I think is the most strategic. Both is pretty strong in like the nontechnical vibe coding space, but it needs to figure out how to enterprise in my personal opinion. There isn't too much content here about how boat can do it. Maybe some stuff here and overall I feel this report is perhaps a little bit too short, right? Like when I do deep research, I want all the information. I want an opinion take, but I want more than just seven pages of content. So now let's look at Gemini's report. And Gemini wrote a much longer report. You see, this is incredibly long report. And it also looked at a ton of different sources. And what I like about Gemini is that it's included some tables about the market sizing. And I think it has a table somewhere here about the different players. right here. So, this is a great table to skim to see all the competitors that Bolt has, what their strategy is, who their primary target audience is, and they did a pretty great job at identifying all of this, right? But let's take a look at Weather is able to actually give some advice on the strategy. So, I'm going to skip to the bottom here. Conclusions, right? Bolt has great tech, strong initial traction, particularly with none developer and vibe coder segments. like this kind of stuff is just like too vague and I don't know it's like gibberish doesn't really mean anything to me. I think Gemini's report even though it's very comprehensive, very long. Some of the conclusions I think could use some work and is a little bit too verbose in my opinion for my liking. So let's take a look at chat GBT. And again, same prompt and it looked at 25 sources, which is the fewer sources, but it did a bunch of searches. And let's see, it has the major players table and their key strengths and weaknesses. And let's look at the conclusion to see what Chat GP came up with. So, I like how Chat GPT summarized the strengths in bold stems with bullets, so it's easy to skim. So speed and magic of results targeting product managers and designers non-engineers in browser tech has a cost advantage and integrations like pursuing integrations and I actually really like this conclusion and there's some pieces missing enterprise strategy or like security but I think this conclusion is actually pretty strong because the speed the targeting nontechnical people the cost advantage and the integrations is exactly what boat is doing. I think of the three deep research reports, I mean, they're all pretty good, but my preference is probably Chat GBT. And the reality is usually when I get a deep research report, I skim through it to understand some takeaways and conclusions. And then what I do is I upload the whole report into an AI project so that I can ask questions for it. So that if I ask AI to produce another document, it has the deep research as a reference. And I think again chat GBT in my opinion is the best middle ground. Claude was a little bit too short and Gemini long and chatbt actually had specific recommendations on strategy based on his research. I think chatb is the winner here. We went a little bit long on this one but deep research is actually really important. It's probably the most gentic like thing that's not coding in AI and I highly encourage all of you to try this. All right. Now

let's wrap up by talking about some AI multimodal capabilities. So, for image generation, my go-to is still Chad GPT. It's great for generating marketing assets, family pictures, and more. Especially if you give it examples of the style that you want, right? So, past some examples of some like marketing asset styles and ask it to tweak the text or replace the image and does a remarkably good job at that. For voice, I think ChachiBT is the winner here. Again, it has the most personality and is the most polished. And you know what most people don't realize is that ChachiT Voice can now sing songs. But the funny thing is it sings them really terribly. And for video, I think Gemini is by far the winner with V3. Being able to go from text to videos with sound is just incredible. And everyone who has Gemini Pro has at least 10 video generations every month. So I highly encourage you to give this a try. If you pay the money, you can generate some pretty amazing short movies and clips. Last but not least, live camera and

computer use are still early features for AI models. Live camera basically means AI can see what's happening on your computer. And I prefer Gemini for this. You have to go to Gemini AI Studio to use this. But for example, you can be vibe coding with live camera on and then you can just ask AI to see if there's any bugs or issues, right? And computer use means that AI can actually take actions on your computer for you by navigating a keyboard and a mouse. And for this one, I prefer OpenAI's operator because it uses 03, which I consider the best reasoning model. So you can use it to look up travel, maybe make restaurant reservations. Again, both of these features are still pretty early, but live camera, I prefer Gemini and computer use, I prefer chat GPT. Okay, so there you have it. So, let's summarize. So, Claude is in my opinion the best at editing, writing, and coding, and that's why it's still my daily workhorse model. ChachiBT is amazing at giving personalized answers thanks to his memory feature. And it also excels in deep research, reasoning, image generation, and voice chat. And Gemini has improved a lot over the past few months. I think it's a great model for coding if cost is important to you, and it video generation feature is mind-blowing and best-in-class. Now, I haven't mentioned Grock at all in this video. They have a bunch of features, but in my opinion, they're not really bestin-class at any of them. But maybe they'll blow us away with their next update. And personally, I still pay for Claude and Chachi PT while dabbling in Gemini once in a while. But honestly, all three models here, Chachi PT, Claude, and Gemini are great. You can't go wrong with either of them. So, there you have it. That is my analysis of the currently the best model for each practical use case. And if you found this video useful, please like and subscribe and I'll make more content like this. See you soon.