Just 3 weeks ago Google released Gemini 3.6 Flash, and now we have 3.7 Flash within 3 weeks. And when I tested 3.6, I was honestly disappointed.

One important thing to notice here: according to Google, this model is based on Gemini 3.6 Flash. So this model is an advancement to Gemini 3.6 Flash, and they’re making some pretty massive claims for better coding, better website generation, better reasoning, and basically many of the exact things I complained about in the last video.

So today, I’m not really interested in the benchmarks a lot. I did some extensive tests with the new 3.7 Flash for web design, for coding, for game development, and I’ll show you what I discovered.

Gemini 3.7 Flash Coding Benchmarks

Just to show you one more thing before we go to the real test, we have these deep space scores here, but I would like to go to the website of Deep and Frontier Code Leaderboard so we can see here where this really stands.

So you can see here, this is somewhere Gemini stands there. So 43.6% scores and the cost of $3.65.

But here, if you see here, very close to it we have other models which are much cheaper than Gemini 3.7 Flash.

For example, you can see Grok 4.6: 48% scores for $2.88 here. You can see Grok 4.5 scores very, very similar – 42.4 for $1.30 only.

And then next to it, we have Kim K3, and then also we have Opus 5 here. You can see it scores very, very high but a little bit more expensive.

Is Gemini 3.7 Flash Really the Best Model for Coding?

So here you can see, for coding standards, it’s not even the best model for coding.

Yes, it’s much improved from the last 3.6 Flash, as this shows, but I don’t think that this is something we expect from Google.

Google is a huge tech company, and we expect something better from Google.

And all those things—releasing pro models, and then changing to flash models, and then releasing in a few weeks the next upgrade models something I really don’t understand where this all is heading.

Gemini 3.7 Flash vs Other AI Models

Yes, they improved their last model, but still, I don’t still think it’s really, really, really that capable.

They need something really big to happen here if they want to stay in this race, if they want to stay relevant.

Okay, so enough talking about all these benchmarks, you can decide yourself.

I did a real test, and then let’s see what the real test says that how good the Gemini 3.7 Flash is.

Gemini 3.7 Flash Real-World Coding Test

So I used Integravity to create one website, to create one game, and to create one app.

These three tests gave me a much better idea about Gemini 3.7 Flash than just looking at benchmark scores.

Test 1: Building an App With Gemini 3.7 Flash

So this is the app I created; it’s a subscription. This app is created using 3.7 Flash.

So they’re saying this is a really fast model I totally agree with it—but the main question is whether the code quality improved or not.

The first version of the app looks really, really nice, but at the same time it’s a typical Gemini app: it created so many things which I never asked it for.

A lot of things are not working.

The Design Looked Good, But It Was Too Complicated

Then I give it like, you know, like “why code it?” I say:

“Okay, you know what, this is too complicated, very overwhelming. Let’s do one thing, make it simple.”

So basically what happened here, it removed the blue color, made everything black, and made everything smaller.

That’s all it did.

Because I told Gemini this is too maximal design, I want something minimal and I want something simple. I do not want all those things.

So what it did, it just changed the colors and made everything smaller. That’s all it did.

If you compare by yourself, not a big difference exists here.

It looks very different, but I really prefer this design. It’s more clear, it’s more minimal than this one; this is a little bit hard to read, a little bit hard to understand.

A Lot of App Functions Were Not Working

Now come to another problem.

I asked it to create, you know, like database and everything so we can have user registration system and everything, but a lot of things are not working.

For example:

• Login button was not working.
• Icons were not working.
• Light mode and dark mode were not working.
• Analytics was broken
• Sometimes data was not saving properly.
• Deleted subscriptions came back.
• Monthly totals were showing incorrectly.
  
  

And this is a little bit not nice, because the same issues I had with 3.6 Flash, that’s why I say it was not the best model.

However, when I ask it to fix the Analytics, it fixed very quickly in one prompt.

But still, we have still a lot of issues.

For example, sometimes when I try to sign in, data is not saving properly. And sometimes when I try to, for example, something I already deleted, I don’t know why it came back.

And here you can see that monthly total is $0.

So here you can see there’s a big mess up.

We already have one subscription, but it says here monthly total zero. I don’t know why it says it’s already like an active subscription.

If I just click on duplicate, now it shows $36, 36.

So these are a lot of issues with this model.

CSV Import Also Did Not Work Properly

I try to work sometime with like, for example, Excel CSV import.

If I choose a subscription, you know, I used with my other app, if I import it, nothing happens here.

And it’s really disappointing because it’s supposed to be a better model.

It’s supposed to be, you know, like a cheaper model.

Of course, it’s a very fast model, but in case of coding, it is really, really a big issue.

I tried to work with a lot of times giving different prompts, but nothing was working properly.

And I really not able to get work done like quickly how I was able to get my work done with Claude, or with GPD 5.6 Soul, or other different, you know, like coding models, or Kim K3 you can say, or Gwen 3.8.

My Verdict on Gemini 3.7 Flash for App Development

So this is the first case.

And I can say is this fast, but it’s okay, it’s a little bit smarter than 3.6 Flash, but I don’t see a huge difference here.

Everybody might have different opinions.

Yes, benchmarks say a lot of things, but for me still, this is not something I will use for this kind of app.

Test 2: Creating a Game With Gemini 3.7 Flash

Okay, next thing I did was creating a game.

There I was a little bit surprised because it did a better job in comparison to its last performance.

I asked it to create a parkour game based in Paris.

It’s not the best game, but yes, it did a better job than when I asked it last time.

We have nice music here, and then it works nice.

All the functions working.

It looks like it’s based on Paris. You can see a very graphical image of, uh, you know, like Paris is there, and it looks fine.

The Game Worked on the First Try

It’s a first try did it, and I was really happy with this game.

Some few things I noticed here which are really nice: so you can choose the map from here where you want to play the game.

But there’s no way to stop the game.

This was a little bit, you know, I was a little bit not happy with this thing.

You can choose the different places where you want to play the game, but the game keeps on running.

And it’s a big thing.

It’s a big thing because I tried this game with other applications also, like other AI models, and this did a really good job in the first attempt.

Of course, the graphics and everything is not, uh, really, really, you know, like something really nice.

And it asked me if I can add a Google Maps API key here, it will be better graphics.

So here I would definitely say it works much better than 3.6 Flash.

Test 3: Gemini 3.7 Flash Screenshot-to-Website Test

Okay, let’s go to the third test when I ask Integravity to create a website using a screenshot.

I already created this video where I explained that how I tested, you know, top five models: Gwen 3.8, Gemini 3.6 Flash, Groq 5% Soul, Opus 5, and GLM 5.2 to create one website using one screenshot.

So I ask the same thing Gemini 3.7 Flash.

I asked it to recreate the website Louis Vuitton.

It looks really, really nice.

But the website created by Gemini 3.7 Flash had the same problem like last time: a lot of templating seems to be done here.

The Website Looks Similar, But It’s Actually Very Different

So here you can see, it looks very similar, but it’s very different here.

For example, it says “Menu” here, it says “Search”, but it should be copying this exact thing here because my prompt was to create an exact similar website.

For testing, it added its own wish list thing because if you create any website with Gemini, it will always add something wish list and this kind of icons.

It’s like a Gemini thing.

So basically, I wanted it to follow this website, but it did not follow this website.

It checked the look and feel of this website, and it added its own kind of thing which looks very similar here.

Is Gemini Using Templates Instead of Following the Screenshot?

In my last video, I explained how this section was created, you know, like looks very similar here, but here we have this button “Quick View” and this “Add to Wish List” button.

They work better, but still, I did not ask it to add all those things because here it don’t exist.

My prompt was clear to exactly copy this website on the right side, but it added own things here.

What it says for me, I understand that it does a lot of, like, it try to understand like what is the design here and the composition here, and it go and search in own database, you know, its own templates, and then it apply the similar looking things here.

That’s why we have this Quick View and the wish list here.

Let me know what do you think about this, why it is happening.

But this is my point of view.

Some Parts Improved Over Gemini 3.6 Flash

Then this here, you can see this is, uh, totally different here.

There is a video here. They added this big banner image here with some text.

Moving on, we have this mess collection here. Very similar.

Last time it did not edit this, uh, button here, it was nothing just this section here.

Now it added.

So I can see it did a little bit better; it saw the design here, so better thing here.

Here we have the wish list thing here on the old website.

So here I can expect that it did a nice job to replicate it.

But again, Quick View was not here, so I don’t know why it added here.

Then this section here looks very similar, and I really think it’s a really nicely done thing.

It also did the same thing in the last test when I did, this is 3.6 Flash.

Iconic handbag, same thing here, same thing here.

The Footer Was One of the Biggest Problems

One thing I was very disappointed was, uh, if I go in the end was the footer section.

This something very badly done.

And one very important thing, it does not copy the—or you can say took the expression from the font—fonts are totally, totally different fonts in the reference website and in the new website.

And then here you can see why I’m saying it’s templates and why it sells like, you know, like visual design at own template.

Because here you can see footer, we do not have this section here where you have this, you know, like this country icon, the currency, and the, you know, client service numbers.

We do not have this here.

But because it thought a footer might look something like this, it added all those things.

In the end you can see we have this, like the country of the delivery here. I do not have it here.

It has here the copyright text, and then you have all these, uh, important links here, and then you can see right the newsletter option here, where on the reference website we do not have this thing.

Why This Kind of Templating Is a Problem

So it might look like it did a good job, but then again, if we are doing the templating and everything, that’s not we want AI for.

Templates already exist in WordPress and other kind of different website builders.

But if you want the AI intelligence, we want something better.

When I did the same job with Claude and ChatGPT and Gwen, they did a very, very good job.

They try to copy everything what was given in the screenshot of the reference website, where Gemini is still lacking behind.

Is Gemini 3.7 Flash Good for Coding?

So yes, you can give all the scores here that how good it is in doing tasks or coding and everything, and for, you know, like all these other benchmarks.

But in my test, it is not standing there.

It might look like it did a really, really good job, but as for me as a developer, I don’t think that it is performing really, really good.

For app development, it was very fast, but I still had a lot of problems with functionality and code quality.

For game creation, I was much more impressed. It did a really good job in the first attempt, and here I would definitely say it works much better than 3.6 Flash.

For screenshot-to-website generation, it looks really nice initially, but when you compare everything properly, there are still a lot of differences and a lot of templating seems to be done here.

So let me know, what do you think about this model?

Have you already tried this model or not?