I Tested Gemini 3 Pro (it did really well)

The Gemini 3 just came out yesterday.

Saw a bunch of people make posts about it. I couldn't help but feel a bit giddy

it. I couldn't help but feel a bit giddy and wanting to give it a try. So, I

basically ran a comparison test between Sonnet 4.5, Gemini 3, and GPT 5.1 High.

So, I must share with you the results, also kind of the pricing they are going to inquire with these and my general thoughts as well. So, let's get into it.

All right. So, this is the first prompt that I tried. All of these prompts uh have been one shots. I didn't want to follow up with different prompts. I just

wanted to see what it can do straight out of the gate. So, all of these tests also are with VIT and React. And the

prompt was essentially create a 3D chess game with three levels of difficulty uh where you can battle against a bot. And

you can also add a scoring system. So,

this is Sonet 4.5 here. And it works. It

works. It got it straight off the bat.

You can move the pieces as you can see.

It works fine. As for the models, uh, they actually look pretty pretty good.

Uh, as you can see the if you zoom in here, right, not too not too shabby. The

only downside, I would say, is that uh the the black pieces here, there's no no lighting on them, so they just look pure black, so it's really difficult to see.

Uh, it hasn't done the scoring system either on it, but at least everything else works. Uh you get the status here

else works. Uh you get the status here whose turn it is as well and you also get the highlights uh which also works really well and as you can see the positioning that you can move. So not

too bad. I'm actually pretty impressed with this. Now let's go over to GPT

with this. Now let's go over to GPT 5.1 high. And this one I didn't really

5.1 high. And this one I didn't really like how it did the models too much on these chest pieces. Uh they all pretty

much look the same. Uh, but other than that, it it's it's pretty close to it.

Uh, I guess the only like plus that it did here is that it keeps track of the timeline. See, if I move the pieces, it

timeline. See, if I move the pieces, it tracks it just here. Uh, so that's pretty nice. But other than that, it

pretty nice. But other than that, it pretty much works the same way. So,

that's kind of the little downside here is that the models just don't look as nice. Okay, let's see what Gemini 3 did.

nice. Okay, let's see what Gemini 3 did.

So, here we go. This is Gemini 3. Uh, as

in terms of the 3D models here that I did, this might be on par if not a bit better than than Sonnet. Definitely

better than GPT.

Let me see if I zoom in here.

Uh, they're pretty much pretty much the same, I'd say. I guess the only pro here I would say is that the dark pieces you

can actually see here because it has proper lighting on it. And also,

as you can see, you have the the letters and the numbers here for the pieces, which I really like. Uh, but other than that, it's pretty much pretty much the same. It did like a circular thing here

same. It did like a circular thing here for your next move as well. Uh, it tried to do the opponent here and kind of the scoreboard, but it gets cut off. So,

that's kind of the the negative, I would say. Uh if I had to pick a winner, uh

say. Uh if I had to pick a winner, uh I'd say GPT here 5.1 is is probably the my least favorite one. And then it's

kind of a neck andneck between Sunonnet 4.5 and uh Gemini. As for the pricing and for the speed, I was actually surprised how long Chant GPT took. So,

Solid and Gemini 3 pretty much took around the same time to do this chess board, which was about 3 minutes and 20 seconds, something like that. Along this

life, uh, I think Sonnet was a slightly faster by like 10 seconds, but pretty much on par. GPD 5.1 high thinking took about five and a half minutes for some

reason, and it generated almost double the amount of tokens to get this project up and running. So, that was a bit surprising to see. Now, the overall cost

for this specific 3D chess game was $1.4 for GPT5. Sonnet was actually 0.5, which

for GPT5. Sonnet was actually 0.5, which was surprising. And then Gemini 3 was

was surprising. And then Gemini 3 was 0.9, so 90. For the next prompt, I wanted to see how well these models do with image detection. How well can they

replicate an image? So, I provided this Nintendo Switch 2, and I essentially wanted all these models to convert it over to an SVG. And optionally, as a

sprinkle sprinkle, they could add an animation by taking these JoyCons and snapping them back into place. So, let's

see the results. So, sonnet, not too bad. I was actually quite surprised by

bad. I was actually quite surprised by this. Um, it didn't get everything

this. Um, it didn't get everything right. For example, I forgot to add the

right. For example, I forgot to add the circles here around the button. As you

can see, they do have those arrows in there. Uh, but it totally forgot the

there. Uh, but it totally forgot the circles. Uh, it totally missed out on

circles. Uh, it totally missed out on the lights here as well and this little portion here that extrudes. It's just

dark. Uh, but other than that, it is pretty pretty close. Now, as for the attachment animation, well, it just moves it and it just translates it and

and it goes behind. So, uh, not the best one, unfortunately. Same prompt for Chad

one, unfortunately. Same prompt for Chad GPT and don't know what to say. I I'm

having such like coin toss results with GPT. It's either really good or it just

GPT. It's either really good or it just terribly falls apart. In this case, it terribly fell apart. That looks nothing like the Switch. The Joy-Cons are

massive. And then the screen is the size

massive. And then the screen is the size of your Nokia phone. Uh, as for the animation and the snapping effect, I mean, it just moves it a little bit

closer. Well, let's see. Gemini, I can't

closer. Well, let's see. Gemini, I can't believe this bloody model. Look at that.

That's probably the most accurate one out of all of them. Uh, yeah, that's the closest it looks to a switch. Now, I'll

give it that it looks more like the first switch rather than the Switch 2, but at least it looks the cleanest out of all of them. Uh, the buttons seem

right. uh this portion seems right and

right. uh this portion seems right and the animation actually properly attaches rather than going behind like we have in

uh Sonnet 4.5. So last place I'd say it's uh it's definitely GPT here. Sonnet

second place and Gemini definitely wins this one. For the next test I did a

this one. For the next test I did a prompt where I wanted it to generate a fitness dashboard here for us where you can track calories, your weight, etc.,

etc. Now, I'm only going to compare Sonnet 4.5 here and Gemini 3 as GPD 5.1 thinking uh it just ran into a bunch of issues and

it didn't even compile. So, I'm going to disqualify it because it is a oneshot test and it failed unfortunately. So, uh

but it's pretty much going to be like Sonnet just a bit more purple if if you can imagine that. That's that's pretty much going to be GPD. Uh but here it is.

Uh, this is Sonnet 4.5. I mean, that's all right. I mean, I don't really like

all right. I mean, I don't really like it at all. And this looks whack. I just

I'm so used to like I can already tell what's AI cards and what's not. And this

just looks like AI cards for me. It did

do recharts here for the graph. So, it's

a good starting point. I'm going to add that to it. Uh, overall pretty decent.

Not too bad at all. Uh, let's check out some other pages. Here we have the workouts. So, I did the filtering

workouts. So, I did the filtering correctly here. The animations are so

correctly here. The animations are so wacky in my opinion. Uh, but not too bad. Then you have the nutrition

bad. Then you have the nutrition tracker. This just didn't load up at

tracker. This just didn't load up at all. Oh, never mind. It does load up.

all. Oh, never mind. It does load up.

Okay, there you go. But you have this big empty space here. Um, okay. It is

what it is. And then you have your progress tracker here as well that uses recharts. So, overall, not too bad.

recharts. So, overall, not too bad.

These animations are just crazy. Uh,

let's see the profile page here as well.

This is a good starting point. I think

if you re refine some of like you don't need an animation on that. Oh my gosh, why why would you do that? Uh, the

layout here is pretty nice though. This

is not too shabby. Um, but yeah, that is Sonic 4.5. Let's see. Uh, Gemini, I must

Sonic 4.5. Let's see. Uh, Gemini, I must say they definitely upped the like the ability to create UI. I feel like with

Gemini 3 uh compared to this use the same icons pretty much. So the cards look pretty much the same here. Uh the

charts here do look a bit nicer in my opinion. Uh the heart rate trend as

opinion. Uh the heart rate trend as well. I do prefer that compared to this.

well. I do prefer that compared to this.

Looks a bit cleaner in my opinion. The

daily activity card is pretty much the same. And they both got this anchor.

same. And they both got this anchor.

Look at that spinning. Jesus. Uh, see

how the line goes? It's just all over the place there. Even on Gemini, it's just not aligned properly at all. Uh,

but let me know which one you prefer. I

do do prefer this a tad bit more. Uh, as

far as let's see the nutrition.

This is how this looks like. If we go to nutrition here, um, I actually might prefer sonnet here.

I do like this water intake like that where you have the little squares rather than having this. And here it doesn't even work. See if I hit plus two 50

even work. See if I hit plus two 50 doesn't work at all. So completely

failed there. Here there's nothing to click on.

I guess you could click on this but nothing happens as well. Uh I do prefer this. Then you have the macro breakdown.

this. Then you have the macro breakdown.

We have the macro breakdown here. This

one looks a bit nicer in my opinion. I

do like the progress lines here though more than than that. Uh but not too bad.

Not too bad. Let's see what else. We

have the weight progress here. Let's

compare it here to this. Um

yeah, overall a bit cleaner. Here you

have these wacky animations. So you do have a bit more content here, but yeah, not too bad. Uh, as far as let's see the

profile here, I must say like the layout here for the profile, I do prefer sonnet over this.

Uh, I guess if you refine this, this could be a bit nicer. Uh, but I probably go with this as the starting point. For

the final result, I wanted to see how creative it can get by giving it just one or two images of a website and trying to recreate it essentially. So I

gave it this front cover here and also a section here like this and then I just told it create all the other pages uh based on the design that this website

has. So use this kind of a template. I

has. So use this kind of a template. I

gave it this image as well. Sorry. So

three images this as well. And then I told it to essentially fill in the rest.

Uh they they do all look quite similar but this is the page I gave it. Well

three images from it. And here are the results. So, we have Sonnet first of

results. So, we have Sonnet first of all and this looks probably the most AIish to me. Uh, if you look at the landing page

me. Uh, if you look at the landing page here, it just screams this is AI. And

the gradients it shows and like the blurring here, it goes orange on top of the green. It just looks weird. And this

the green. It just looks weird. And this

just looks super amateur-ish to me. Uh

as you can see just like tried to recreate it but complete completely failed. Uh there's no animations as

failed. Uh there's no animations as well. Uh no there is a bit of animation

well. Uh no there is a bit of animation here. I guess this fading out but wasn't

here. I guess this fading out but wasn't too too impressed with this. And then

they used emojis here for the trusted brands because that's what you want to see. You want to see that little purse

see. You want to see that little purse emoji on Chanel. My gosh. And next up we have Gemini here. And as you can see, this actually tried to get the

aesthetic, which I really appreciate. As

you can see, the font as well looks quite close to it and all these little elements. Now, do I like this? No, not

elements. Now, do I like this? No, not

really. Uh, but it's it's a good start.

I'd say that. And you got these animating as well, these little uh badges, which is pretty cool. And look

at the little uh navigation here at the bottom. That's pretty much uh

bottom. That's pretty much uh they got pretty close there. I'm not

gonna lie. If you look at Sonnet, it doesn't even have it at all. So, not too bad. Like the nav as well, how they did

bad. Like the nav as well, how they did it. Much closer to the original image

it. Much closer to the original image compared to to Sonnet. Um, not too bad.

Then we scroll down. We actually have some pretty decent looking animations here. As you can see, it did a pretty

here. As you can see, it did a pretty good job of recreating it from the the photo I gave it. So you have the animating cursors here and you also have this little animation where the circle

starts expanding. Let me refresh that

starts expanding. Let me refresh that for you really quickly. There we go. So

not too bad. And then we got this section here as well. Uh we got these cards which look pretty all right. Not

too bad. Nothing too great compared to the original that I provided. Of course,

uh there's this section here, right?

This is a bit more uh refined, if I could say that. Uh, but

not too bad. It tried. Uh, pretty

impressed with it. It didn't do any other pages, so it got a bit lazy. Uh,

but definitely more impressed with this compared to sonnet. As for chat GPT, again, the landing here, I still think

this is the best so far. If you look at this and then you look at this, but it did a far better job than Sonnet. And I

actually prefer uh some of the cards it did uh here even compared to Gemini. So

here, this one doesn't look too bad, honestly. Uh it did a pretty good job.

honestly. Uh it did a pretty good job.

Got the floating badges here. The

trusted brands here, it actually animates. So you got to give it some

animates. So you got to give it some props there as well. Let me make this a bit bigger here for you. There we go. So

that's that's really good. Card section

here looks all right. This looks all right. And that's pretty much everything

right. And that's pretty much everything that it did. So

tough. But again, if I have to say Sonnet definitely is the last place here. It's this is terrible. I would not

here. It's this is terrible. I would not use any portion of this GPT.

Don't like the the homepage here, but I would maybe consider actually taking portions of it. This is a good start here. This looks almost there. Maybe 80%

here. This looks almost there. Maybe 80%

there. This whole trusted brands ad.

Slap some icons on there and you're good to go. Uh everything else is is meh, not

to go. Uh everything else is is meh, not too good. And Gemini,

too good. And Gemini, again, you can argue that this is like closer to the actual OG website here

with everything that I tried to do. Uh

and the landing page, I feel like here is as well a pretty good start. If you

just kind of reposition some of the elements, add a bit of finesse to it, uh you are there. So, I'll let you guys uh kind of judge yourself. Uh but again,

this is kind of an early test. These are

the results I got. Uh let me know if you want to see more. I can definitely do a followup even with more models if you prefer. But yeah, really fun, really

prefer. But yeah, really fun, really interesting stuff. And crazy how this uh

interesting stuff. And crazy how this uh damn. I honestly thought that it just

damn. I honestly thought that it just felt in the last like year or so that we're kind of stagnating with the progress of AI models. Uh but with

Gemini 3, I do feel like okay, we are getting a bit of a step up. Uh so it's just a welcome change. Let me know what you think. Catch you guys in the next

you think. Catch you guys in the next