It seems like the current default path

is heading towards a very, very scary

place.

>> DeepSeek

>> DeepSeek

>> DeepSeek

>> And it is shaking this entire industry

to its core.

>> be really bad for China to be ahead in

AI capabilities.

>> China is beginning to engage in what

I'll call AI dumping.

>> About a third of corporations now are

supposedly using Chinese lightweight

open weight models that are cheaper.

>> If I were Xi, I would just dump cheap AI

into the US market. I think the US

market crashes.

>> They practically gave the whole thing

away for free.

>> It might be over for Claude and ChatGPT.

>> DeepSeek just released their own

harness, and this changes everything we

know about coding agents. It can do

everything that Claude code can do. It

can run commands, read files, but you

can use any model you want inside of

this harness. But do you know what's

very different? The price. I'm paying

$200 a month right now for Claude code

max.

>> Cursor is 20 bucks, but this new

DeepSeek harness is zero. Yes, you heard

that right. No free trial, not a starter

tier.

>> Zero. All you do is that you bring your

own API key.

>> There are a lot of people right now that

are making guides on YouTube right now

that are saying that you can stop your

Claude code subscription. However, you

should know by now that most setup

guides, they don't tell you the entire

truth. Because if you watch what happens

in those videos, they set it up, they

build some simple project, they say it

works, and then they're done. But no one

actually shows you the bills that break,

the things that actually matter, and the

thing that determines if you're going to

switch or not.

>> So, is DeepSeek harness actually as good

as people say it is? Well, I ran the

ultimate test against both Claude code

and Codex. Let's not waste any more

time. Let's get into it. Let's start

with the absolute most important thing

first, which of course is the pricing.

>> In order to have access to the Codex app

and use it for actually building stuff,

you would need the $20 a month

subscription, so ChatGPT plus, because

this gives the expanded access to Codex,

which you would need. The $8 a month

doesn't give you enough usage. It allows

you to build with the Codex app for

macOS and Windows. It has access to the

advanced reasoning models, which are the

ones that we're going to be using. These

things are pretty crucial if you're

serious about building anything. So,

Codex is $20 a month. Clo is writing it

in a little bit of a sketchy way, in my

opinion. They're writing $17 a month

right here. This is only if you take the

annual subscription. It is also $20 a

month when you pay monthly, which most

of us do. The problem with the Pro plan

is that you probably already know it. We

don't have that much usage on the Pro

plan. So, if you actually want to build

real stuff for your business, then you

would probably need the Max plan, which

is $100 a month. So, Clo is definitely

the expensive option. And then we have

the DeepSeek Harness, which was

literally just released. And the crazy

thing about this is that it's fully open

source. You can see, I can find the

GitHub repo right here. It has 177,000

stars already and almost 20,000 forks.

And the entire harness is open source,

which is pretty cool in my opinion.

However, one thing that we need to take

into consideration when we do the test

is that this is still in preview. They

literally just released this as a

preview to what is coming. And we're

going to talk more about this, but the

main thing that I think is super cool

about the DeepSeek Harness and is a

massive opportunity is the fact that

every capability is a plugin, they call

it, that can be swapped or recomposed.

So, we can literally change anything we

want about this harness. We can

customize it to our exact use case,

which is something we cannot do with

Codex or with Clo code, because that's

of course closed source. In this video,

I'll show you some insane use cases and

we're going to put it to the ultimate

test to see if this is just another hype

video or if it's actually something

that's good. Let's get into it. There

are multiple ways that you can install

the harness, and a lot of them are

pretty confusing and pretty technical,

but I'll show you the absolute easiest

way. Just copy this and paste it in,

say, install this, and it's going to

install it in a couple of minutes.

However, if you don't have Clo or Codex

installed, I've created this

installation guide, so you can just

follow this with these commands. I'm

basically just going to create a new

folder. I'm going to call it DeepSeek.

Then, I'm going to open up Visual Studio

Code. If you don't have it already, you

can just download it off the internet

just by searching Visual Studio Code

download, click open, and then choose

the DeepSeek folder that we just

created, just like this. Open the

terminal, and now we are ready. All we

have to do is just paste in this command

right here and hit enter. It's in the

installation docs. And this is going to

download and set up the harness for us.

When it's done, it's going to give you

this URL right here where the harness is

open. So, just paste that into your

browser, and now we are inside the brand

new harness that is still in preview.

There's a couple of things that we're

going to do before our test. The first

thing we're going to do is that we're

going to set it to full access. And for

the model, we're going to choose the

DeepSeek V4 Pro, which is the strongest

DeepSeek model. It's a lot cheaper than

the other models. I gave the pricing

from the Cloud API and the pricing from

the DeepSeek API. And what we'll be able

to see is that 5.6 Soul is 15 times more

expensive when it comes to the output

and 7 and 1/2 times more expensive when

it comes to the input. Opus is almost

the same as GPT 5.6 Soul. And then Fable

5 is 15 times more expensive on the

input and 25 times more expensive on the

output. So, if the DeepSeek V4 Pro is

just close to being as good as the other

models and can perform something that's

close, I would say that's a win for

DeepSeek. But, let's see. This is my

brutally honest test. We're actually

going to be starting with the hardest

and most exciting test first, which is

going to be running for a while. I want

all the models to try and build in Loom

clone with full authentication, full

back end, full database, absolutely

everything. So, what I'm going to do is

that I'm going to copy this entire

thing, and then I'm going to paste it

in. I'm going to set it to standard

mode, which is the full coding agent

with file editing and everything. We're

going to set it to the DeepSeek V4 Pro,

and we are going to set it to effort

high. So, this is what we're going to do

on all of them. I'm going to set it to

full access, so I don't have to sit

there and accept everything that it

does. And then we run it. And then we're

going to do the exact same for Claude.

We're going to paste in the exact same

prompt. We're going to use Opus 5, which

is the closest equivalent to the

DeepSeek model. It is still like 10

times the price. But again, what we're

testing right here is not necessarily

the model, it's the model and the

harness together. We set it to auto mode

and then we hit enter. And then we're

going to take the exact same prompt,

give that to Codex as well, use 5.6 with

high, and then run that prompt as well.

And there we have it. All three

harnesses are now running side by side

on the exact same prompt. I can see I

just forgot that Opus 5 of course needs

to be on high as well, same as the other

models, so it is fair. But now my guess

is that it's going to take a couple of

hours for each models now build this

out. So I'm very excited to see the

results. In the meantime, I want to show

you some extremely cool things you can

do inside of DeepSeek as well. That's

pretty unique to a harness that I've

actually never seen before. DeepSeek has

done something that's very cool that

I've never seen before inside of a

harness. Usually you're only able to see

the chat. You're not able to see what's

actually going on behind the scenes. But

we have this button right here,

trajectory, where we can literally see

all the tool calls, what the AI is

thinking, exactly what's going on

throughout this entire process. Because

it is open source, it allows us to

actually look behind what's going on,

which is extremely cool. And on top of

that, the customization that I talked

about earlier, we can go inside of

settings and then we can go inside of

plugins. And then inside of the plugin

list, we can literally enable and

disable every single tool inside of

here, which is pretty cool, not going to

lie. And because it's open source as

well, and we are running this harness

locally, we can actually update and tune

this to whatever we want. It's almost

like playing with Legos. All right, all

harnesses are running side by side. I'm

extremely excited to see the result of

how well this new harness performs

compared to Codex and Claude. Because if

it's even close, and we are paying 1/10

of the price, that would be such a

game-changer. Well, only time can tell.

This is going to take a couple of hours

before they are ready. Now comes the

part to I look at the result for what

these models and harnesses actually

built. We have three separate platforms

that look kind of similar. Some are

definitely better than other and I can't

even remember what platforms are which.

So, I'm going to look at them, give my

honest opinion about what platform I

think is best and then you'll see

exactly what harness and model is

actually best. So far, just looking at

this landing page right here, this looks

extremely good and clean. I really like

how it is designed. This honestly looks

very good. The pricing card they look

clean. It does look a bit vibe coded so

you will still need to put in some

effort when it comes to like making this

animation look not as vibe coded. But as

a landing page off of one prompt, I

think that's pretty good. Let's log in.

Already now running into a bug for the

first software. I'm going to fix the bug

so we can see the rest of the platform.

We just need to keep in mind that we

already have a bug on this platform. We

can't even log in. Okay, to be fair, it

wasn't an issue with the software. It

was just the port on our computer that

we were running it from. Let me log in

right quick. We receive an email like

this where we can just click open. This

email, to be fair, looks very good and

professional. So, that's pretty good.

And then we actually have an onboarding

that looks good. Start my first

recording. It asked us for our camera

right here. That's pretty cool. Let's

click allow this time. Allow for the

microphone and let's click start

recording. Entire screen. 1 2 3 testing.

1 2 3 testing. Is it working like it

should? This screen right here that we

just saw, that was a bit buggy though.

So, not the best user experience there.

Recording didn't start. Bug number two

that we need to fix. All right. Now it

actually uploaded this time. Let's see

if it actually works. It does. The

recording is kind of messed up but this

is, to be fair, what I saw on my screen.

So, now we have a fully functioning Loom

right here. Pretty impressive, not going

to lie. And it has the link that we can

send to people to share. This issue

right here with the images though, but

like we have a fully functioning app and

the UI honestly looks very, very good.

We even have multiple workspaces. And we

can even create folders. Not bad for the

first one. Let's check the other two.

This is Loomy. I actually have no idea.

I forgot which one was which. But we can

sign in here as well. We get this login

link right here where we can sign in

with a magic link. All right, we have an

onboarding again. Product demos,

continue to our first recording. There

we go. Allow, default camera, default

microphone. Not showing a preview right

here. Start recording. Hello, does this

work? Can we see the screen right here?

So, I bug with this software as well.

Let me try and fix that so we can have

three working products so we can compare

them one by one, but of course we got to

keep in mind how much time it actually

takes to fix it because so far none of

these harnesses has actually one shot

this. All right, fixed the first bug in

this software so we can click start

recording, choose our entire screen.

This is going to start the countdown and

I can say yo, yo, yo. Can you hear what

I'm saying? We can show this screen a

little bit. We can show this and then we

can stop the recording. Finalizing your

recording. That was actually pretty

quick this time. Paste this in and see

if it works. It does. And it shows me

this little squish though, but we have a

fully working Loop Count with audio. All

right, not bad. It doesn't have the

entire platform like the other one had,

so that's a bit of an L, but it can do

what we need it to do. And then we have

the last one also called Loomy. We can

also sign in here. The UI looks pretty

good here, not going to lie. Then we

have this like an email right here where

we can like create an account. We have a

very clean onboarding. This looks good

and then we can like start recording to

our screen. Let's see if it can actually

one shot this cuz none of these has been

able to one shot it so far, but it would

be cool if one of them can one shot it.

So, my like screen is down here in the

left hand corner. So, it's showing while

it's recording. Go a bit to this screen

as well. Go back a little bit and click

stop. Wrapping up your recording. Upload

problem once again. So, every single one

of the harnesses couldn't one shot this,

which is very, very interesting. Let me

fix that and be right back. All right,

and there we have it. Now the upload

works. Exact same bug as the other one,

so it's very interesting how no one

caught that one. But now we have a

perfect screen recording software. And

this software also has a my library tab,

but that doesn't work.

Let me try and sign into that again and

see if it works then. Cool, so this

platform also has an entire back end

that looks pretty good. I like this.

It's very intuitive. So, that definitely

gives an extra point as well. All right,

the time for the big reveal. What

harness and model actually created each

of these? The first one that had a code

of box was actually Codex that worked

for an hour and 46 minutes. So, it took

almost 2 hours to build up this

platform. And I'm going to say based on

the free we have right here, I like this

UI the best. It's nicer on the eye. This

looks very by coded, however, but like

in general, the platform looks nice.

Number two that's running on port 3100

was actually DeepSeek and you can see it

has run for 82 minutes. So, what is

that? Like an hour and 20 minutes? That

is not bad. Considering this model is

1/10 of the cost of the two other models

that we're using right now, it creating

this and creating a working product, we

had one bug we had to fix. It's honestly

really, really impressive. I still like

the UI better on this one. I like the

platform and how intuitive it was

better, but considering that it's an

open source model and that it's 1/10 of

the price, I'm honestly extremely

impressed. I didn't expect that. And

that also means the last one, this one

right here, the third one was made by

Claude Code. The crazy thing is that

Claude Code and Opus was the one that

took the longest by far. It took 3 hours

and 27 minutes to build out, which was

twice as long as DeepSeek and almost an

hour more than Codex. But what do you

guys think? Let me know in the comments

which UI, which logic that you like the

best and if you're going to be using the

DeepSeek harness yourself. If you want a

full guide for how you can take your AI

knowledge and turn it into a business,

then you can consider our community. You

can try it completely for free. It has

all of my Claude skills and it has the

7-day challenge that takes you from

nothing to building an AI business in 7

days. I'll leave a link in the

description below. It's safe to say that

Chinese AI is catching up. I'll say that

I like this one the best from Codex,

then probably Claude, and then DeepSeek.

But DeepSeek is not far from the other

models and it's 1/10 of the price. Thank

you so much for watching. Albert out.