AI Coding Battle: DeepSeek vs GPT-5
45sThe opening sets up a direct, high-stakes comparison between two popular AI models, immediately engaging viewers interested in AI and coding.
βΆ Play ClipThis video pits DeepSeek V3.1 against GPT-5 Pro in a series of coding challenges, testing their ability to generate functional front-end projects. Over seven prompts, the creator evaluates each model's output quality, speed, and feature accuracy, ultimately declaring a winner based on a point system.
The video compares DeepSeek V3.1 Deepthink and GPT5 Pro in a coding battle, focusing solely on coding benchmarks.
GPT-5 Pro's version worked perfectly with accurate collision detection and sound effects, while DeepSeek's had a missing pipe gap and faulty pixel detection.
DeepSeek's snake game worked but had bugs causing repeated game-over. GPT-5 Pro's version was completely non-functional, with buttons unresponsive and the game starting prematurely.
Both models produced good results, but the creator preferred GPT-5 Pro's neon theme and overall design, giving it the point.
GPT-5 Pro's website was fully functional with working filtering, quick view, and add-to-cart. DeepSeek's version was mostly non-functional 'showpiece'.
DeepSeek excelled with a real-time updating UI, despite a minor scrolling bug. GPT-5 Pro's dashboard had a terrible UI and no working functions.
DeepSeek outpaced GPT-5 Pro with a well-designed site featuring price alerts, portfolio, and candlestick charts. GPT-5 Pro's output was unfinished and disorganized.
GPT-5 Pro's version had better visuals, clickable planets with info, and asteroid layers. DeepSeek's was decent but had non-working clickable planets and strange planet visuals.
GPT-5 Pro narrowly wins the coding battle by one point, but DeepSeek V3.1 offers excellent value and competes closely, often excelling in specific tasks like real-time data dashboards. The creator suggests both models have strengths, but for this benchmark, GPT-5 Pro takes the win.
"The title accurately promises a coding showdown between the two models, and the video delivers exactly that with detailed tests and a clear winner."
Which model won the Flappy Bird clone test?
GPT-5 Pro.
1:56
What bug did DeepSeek's Flappy Bird have?
No gap between pipes and faulty pixel detection causing repeated game over.
1:26
How many points did GPT-5 Pro win by?
One point.
10:34
In which test did DeepSeek outperform GPT-5 Pro?
Business Dashboard (Test 5) and CryptoTracking Website (Test 6).
7:30
What was the final score of the competition?
GPT-5 Pro won by one point, with both models ending at 3-3 before the final test.
9:12
What feature did GPT-5 Pro's 3D solar system have that DeepSeek's lacked?
Clickable planets with information panels and asteroid layers.
10:21
What was the creator's overall verdict on both models?
Both are great; DeepSeek is more cost-effective, but GPT-5 Pro won this coding benchmark.
10:39
GPT-5 Pro wins Flappy Bird Test
Shows GPT-5 Pro's ability to generate a fully functional game with accurate collision detection and sound effects.
1:56DeepSeek wins Snake Game
DeepSeek's game was playable despite bugs, while GPT-5 Pro's was completely broken, demonstrating that older models can sometimes outperform newer ones.
3:29GPT-5 Pro's Design Preference
Showcases that subjective UI/UX can influence which model is preferred, even when both meet functional requirements.
5:17GPT-5 Pro's Functional Shopping Site
Demonstrates GPT-5 Pro's strength in creating complex, functional e-commerce features like filtering and cart management.
6:54Final Score Ties Before Last Test
Highlights the intense competition and that neither model dominates consistently across all tasks.
9:12[00:00] Hi guys, welcome back to the channel. In
[00:02] today's video, we're going to show you a
[00:03] comparison between Deepseek 3.1 versus
[00:07] GPT 5 Pro model. In our previous video,
[00:10] one of our subscribers commented asking
[00:12] us to create a video on GPT Pro versus
[00:15] Deepseek 3.1. So, here we are with
[00:18] another brand new video. So, let's jump
[00:21] straight into the action. Setting up our
[00:23] coding battle.
[00:25] So here as you can see we have the GPT5
[00:28] Pro model and on the other hand we have
[00:31] DeepSeek V3.1 Deepthink.
[00:35] In this video we're purely going to
[00:36] focus on coding benchmarks. So this is
[00:39] going to be intense.
[00:43] >> Here comes our first prompt. Generate a
[00:46] complete Flappy Bird clone using HTML
[00:49] CSS. include pixel perfect collision
[00:51] detection, smooth animation, score
[00:54] tracking, and sound effects using web
[00:55] audio API. Let's give this prompt to
[00:58] both models and let's see what happens.
[01:00] So, as you can see, DeepSeek started his
[01:02] thinking and GPT Pro is thinking, too.
[01:07] And to my surprise, Deepseek finished
[01:09] his thinking and also started writing
[01:11] code.
[01:13] Sometimes it's fast, too. That's
[01:14] interesting. Okay, now Deepseek is
[01:17] finished. Let's check the code and
[01:19] Here's the game.
[01:22] >> Let's click on start.
[01:24] >> Guys, what the heck? How am I supposed
[01:26] to cross this pipe if there's no gap?
[01:29] >> And also, pixel detection is not
[01:30] working. It's repeatedly showing game
[01:33] over.
[01:35] >> This is not good, Deep Seek. Now, let's
[01:37] check GPT Pro. So, here's the finished
[01:40] code. Let me copy the code and test it.
[01:42] Let's click on start. And yeah, it's
[01:46] working. It's perfectly working. The
[01:48] pipe detection is working accurately.
[01:50] Score function is working too. And it
[01:52] has nice sound effects. Not harsh, not
[01:54] too subtle.
[01:56] >> So overall, it's working perfectly. One
[01:59] point to GPT5 Pro.
[02:06] [Music]
[02:11] >> Test two, snake game challenge. Now,
[02:14] here's our second prompt. Create a
[02:16] modern snake game with neon glowing
[02:18] effects, dynamic grid scaling, multiple
[02:21] difficulty levels, food powerups, smooth
[02:23] snake movement, and gameover animations.
[02:27] Let's send this prompt to both GPT as
[02:30] well as Deepseek model.
[02:32] Okay, so here Deepseek is finished. Let
[02:35] me copy this code and test it. Let's
[02:37] click on start game. And it seems it's
[02:39] not working. Let me click on easy level.
[02:43] Okay, it started working. I think there
[02:45] are some function misalignments, but
[02:47] anyway, let's play this game. Okay, did
[02:49] you guys notice why we got game over
[02:53] H? Okay, I got it. We're getting the
[02:56] same game over again and again. There
[02:59] are some errors and bugs in the code.
[03:01] So, DeepS works, but not fully
[03:03] perfectly. It needs to fix some bugs and
[03:05] then only it can run properly.
[03:08] Now, let's check GPT 5 Pro. So, here's
[03:12] the GPT output. Let me copy the code and
[03:15] test it. We got game over before we even
[03:18] started the game.
[03:20] GPT, you really messed up. Terrible
[03:24] mistake. No buttons are working and the
[03:26] game already started before we even
[03:27] pressed start.
[03:29] >> Deepseek gets one point. At least DeepS
[03:31] was working. Now both models have equal
[03:34] points. Let's move to the next prompt.
[03:38] Generate a cuttingedge portfolio website
[03:40] for a freelance video editor featuring
[03:42] video reels. Showcase with custom video
[03:45] player, before after project
[03:47] comparisons, client testimonials,
[03:50] carousel, timeline of projects, contact
[03:53] form with animation, and parallax
[03:56] scrolling effects.
[03:58] Let's send this prompt to each model.
[03:59] And I'm noticing that today Deep Seek's
[04:01] thinking is very fast. Like it's hardly
[04:04] taking minutes and starting the code. So
[04:07] that's a good point. There's definitely
[04:09] improvement in coding. I can surely say
[04:11] that about Deep Seek. Okay, so Deepseek
[04:14] is finished. Let me copy the code and
[04:16] test it. Here's the website created by
[04:19] DeepSeek model. Looking good. The color
[04:22] theme, animation effects, and it has all
[04:25] the requirements that meet our
[04:26] expectations.
[04:28] There are client testimonials as we
[04:30] asked for and project timeline as well.
[04:33] So, it accurately followed our prompt so
[04:35] well.
[04:37] Good job, DeepSeek. Now, let's test GPT5
[04:40] Pro.
[04:42] Okay, copy the code and let's see the
[04:43] output.
[04:45] >> Okay, guys, first reaction. Wow, the
[04:48] color theme picked by GPT is amazing.
[04:50] Like this blue neon theme giving that
[04:52] tech vibe like it should as this website
[04:55] is designed for a video editor.
[04:58] So, the color theme is excellent and
[05:00] there are video samples which are empty,
[05:02] but they're there. And the project
[05:03] timeline, the project timeline looks
[05:06] good to me. If you look closer, there's
[05:08] slight misalignment. It should be at the
[05:10] center of the line, but anyway. So,
[05:12] there's a contact form with slight
[05:14] animation. Overall, I like the GPT
[05:17] website more than Deepseek. It could be
[05:20] a personal choice, too. Some of you may
[05:22] like the Deepseek website, but for me
[05:24] personally, I'm liking the GPT website.
[05:27] So, GPT gets one more point. Now, GPT is
[05:30] leading in our comparison. Let's move to
[05:32] the next prompt.
[05:35] Test four. Create a high-end shopping
[05:38] website with product filtering system,
[05:40] interactive product viewer, shopping
[05:43] cart with slide out animation, wish list
[05:45] functionality, product zoom on hover,
[05:48] rating system, and checkout process with
[05:50] form validation. You can see we're
[05:52] giving hard and specific functions so
[05:54] that we can see who follows the prompt
[05:56] more accurately.
[05:58] Let's send this prompt to both models.
[06:00] And you can notice DeepSk's thinking is
[06:02] very fast today.
[06:04] I noticed that in coding, especially
[06:06] DeepSeek improved its thinking. Now it's
[06:08] more efficient and faster than the
[06:10] earlier version. Okay, now Deepseek is
[06:13] finished. Let me copy the code and test
[06:15] it. And here's the DeepSseek output. The
[06:19] theme is very generic, very basic
[06:21] website, and product filtering functions
[06:23] are not working very well. None of the
[06:26] buttons are working, so it's just a
[06:27] showpiece, nothing else. DeepSeek
[06:29] disappointed us here. Anyway, let's
[06:32] check GPT5 Pro output. Okay, so here's
[06:35] the code. Let me copy this and check.
[06:39] And now this is the website generated by
[06:41] GPT5 Pro. And I must say this looks
[06:43] good. No joke. The filtering functions
[06:47] are working perfectly. Quick view is
[06:49] also working and add to cart is fully
[06:52] functional.
[06:54] >> So GPT really put some effort here and
[06:56] made it functional as well as we wanted.
[06:58] So one point to GPT again.
[07:01] Now GPT is two points ahead of Deepseek.
[07:04] Let's give our next prompt.
[07:08] >> Test five. Generate a comprehensive
[07:10] business dashboard showing AI
[07:12] development metrics, profit analysis,
[07:15] growth charts using chartjs, animated
[07:18] counters, and real time data simulation.
[07:20] Okay, this is a bit harder than the
[07:22] earlier ones. Now let's send this prompt
[07:24] to both models. While deepseek is
[07:26] finished, let's test it.
[07:30] Okay, this time DeepSeek excelled at
[07:32] this. It's updating in real time and the
[07:35] UI is looking good to me, but there's a
[07:37] bug I noticed. It keeps updating in real
[07:39] time, but there's no function to handle
[07:41] long earlier data, which is making the
[07:43] website length keep going on and on. But
[07:46] overall, this website is looking good.
[07:49] Now, let's test GPT 5 Pro output. And
[07:52] here's the output. And GPT again messed
[07:55] up this. The UI looks terrible with no
[07:58] functions at all. There's nothing to
[07:59] watch. Okay, let's give the point to
[08:01] DeepSeek. Now, Deepseek is only one
[08:04] point behind GPT. What do you think,
[08:07] guys? Will Deepseek overtake GPT5 Pro
[08:10] here? Comment down below your
[08:12] predictions, and those who accurately
[08:14] predict will get a shout out in the next
[08:16] upcoming video. Test six, cryptotracking
[08:19] website. Okay, so here's our next
[08:20] prompt. Build a cryptotracking website
[08:22] with real-time price updates,
[08:24] interactive candlestick charts,
[08:26] portfolio calculator, price alert
[08:29] system, and animated price change
[08:31] indicators with color coding. Let's send
[08:33] this prompt to Deepseek as well as GPT5
[08:36] Pro. Okay, Deepseek is finished. Let me
[08:40] copy the code and run this. And here's
[08:42] the output. It's looking good with all
[08:44] the functions we expected. There are
[08:46] price alerts and portfolio and
[08:48] candlestick charts. And the UI is
[08:50] overall looking good. Now let's test GPT
[08:53] 5 Pro. Okay, so copy this code and
[08:56] here's the output.
[08:58] GPT, what happened to you, boy? This is
[09:00] what they call
[09:02] >> Well, let's just say it looks very bad
[09:04] and unorganized and it's not even
[09:06] finished properly. Deep Seek again. Got
[09:08] this point. Now Deepseek and GPT are at
[09:12] the same point. Now let's give the final
[09:14] prompt and we'll know who will be the
[09:16] winner of this video. Deepseek 5 3.1 or
[09:20] GPT 5 Pro. Final test 3D solar system.
[09:24] Here's our last prompt. Generate an
[09:26] interactive 3D solar system using CSS 3D
[09:30] transforms in JavaScript. Include
[09:32] realistic planet orbits, clickable
[09:34] planets with information panels, zoom in
[09:36] and zoom out functionality with planet
[09:38] rotation animation. Let's give this
[09:40] prompt to both models. Okay, Deep Seek
[09:42] is finished. Let me copy the code and
[09:45] test it. And here's the output.
[09:48] The sun looks good, but the planets are
[09:50] kind of strange.
[09:51] We have zoom in andout features as well
[09:54] as a reset button. And the planet
[09:56] clickable function is not working, but
[09:58] overall it's looking decent.
[10:01] Now, let's test GPT5 Pro Code. Let me
[10:06] copy the code. And here's the output.
[10:09] This looks much better than DeepSeek for
[10:11] sure. The sun looks good here. and
[10:13] planet visuals are better than deepseek.
[10:16] It also has zoom in andout features and
[10:18] has asteroids in the outer layers. And
[10:21] one more thing this has is the planet
[10:23] clickable function which shows the
[10:24] information of that planet when you
[10:26] hover over it.
[10:28] So guys, I think GPT5 Pro clearly gets
[10:31] the point here. Final verdict and
[10:34] conclusion, GPT5 Pro model wins this
[10:37] test by one point. Deepseek is very
[10:40] cost- effective and in terms of value
[10:41] for money and effectiveness, Deepseek
[10:43] will win. But in this pure coding test,
[10:46] GPT won this time. Both models are great
[10:48] at coding and sometimes Deepseek is
[10:50] better than GPD5 Pro. But if you really
[10:53] ask me who's the winner, then I would
[10:55] say both are winners. But for today's
[10:57] video, in terms of benchmarking, GPT5
[11:01] Pro is the winner of this video. GPT5
[11:04] Pro sometimes gives exceptional
[11:07] front-end code that looks very good
[11:08] compared to Deepseek's basic generic
[11:11] websites that look very plain sometimes.
[11:13] For the next video, do comment below
[11:16] which model you want to test next. I'm
[11:18] thinking of bringing Claude 4.1 Opus
[11:21] thinking model and Kimmy model 2 for our
[11:23] next videos. Do let me know your
[11:25] thoughts. On this note, let's end this
[11:27] video here. If you haven't subscribed
[11:30] already, then subscribe to our channel.
[11:32] Many exciting videos are coming very
[11:34] soon. Thanks for watching and I'll see
[11:36] you in the next one.
[11:41] >> Through the wastelands, through the
[11:43] highways, till my shadow turns the sun
[11:46] raise. And on we
[11:51] [Music]
β‘ Saved you time reading this? Transcribe any YouTube video for free β no signup needed.