---
title: 'I Tested DeepSeek V4 vs Opus 4.7 vs GPT 5.5'
source: 'https://youtube.com/watch?v=uT2m7VD99qA'
video_id: 'uT2m7VD99qA'
date: 2026-06-15
duration_sec: 1620
---

# I Tested DeepSeek V4 vs Opus 4.7 vs GPT 5.5

> Source: [I Tested DeepSeek V4 vs Opus 4.7 vs GPT 5.5](https://youtube.com/watch?v=uT2m7VD99qA)

## Summary

The video compares three major AI models—GPT 5.5, Opus 4.7, and DeepSeek V4—released within 24 hours, testing them on coding tasks to determine which is best for users. The host evaluates cost, benchmarks, and real-world performance in building a flight simulator and a WebGPU landing page.

### Key Points

- **Cost Comparison** [00:58] — DeepSeek V4 is 8x cheaper than competitors: $30 per million output tokens for GPT 5.5, $25 for Opus 4.7, and $3.48 for DeepSeek V4. Input tokens: $5 for GPT/Opus, $1.70 for DeepSeek.
- **Benchmark Results** [02:40] — On SWE-bench Verified and SWE-bench Pro, Opus 4.7 wins. On Terminal Bench 2.0, GPT 5.5 leads with 87.2%, beating Claude's Mythos. DeepSeek V4 is third but within 5 points of Opus while being 8x cheaper.
- **Long Context Performance** [03:49] — Opus 4.7 performs poorly on long context (500K-1M tokens), worse than DeepSeek and GPT 5.5. GPT 5.5 and DeepSeek handle long context better.
- **Flight Simulator Test** [05:10] — GPT 5.5 (Codeex) produced the best flight simulator after 3 iterations (66K tokens, ~15 min). Opus 4.7 (Claude Code) was second but needed more prompts (200K tokens, ~20 min). DeepSeek V4 (Open Code) failed completely.
- **WebGPU Landing Page Test** [17:07] — Opus 4.7 created a subtle, tasteful design (175K tokens). GPT 5.5 was flashier but overly bright (107K tokens). DeepSeek V4 produced a seizure-inducing, low-quality result (130K tokens).
- **Final Verdict** [23:12] — GPT 5.5 wins for complex tasks (flight sim). Opus 4.7 wins for design taste (WebGPU page). DeepSeek V4 is only suitable for simple, budget-conscious tasks.

### Conclusion

GPT 5.5 and Opus 4.7 are both strong choices for agentic coding, with personal preference playing a big role. DeepSeek V4 is a budget option for simpler tasks but lags significantly in quality.

## Transcript

In the last 24 hours, we have had huge
updates to two of the biggest AI models
on the planet. First, we got the release
of GPT 5.5, which is boasting certain
benchmark scores that beat out Claude's
Mythos. Secondly, we got the release of
DeepSeek V4, which is an open-source
openweight model that has benchmarks
that rival these frontier big players.
So, with all these new models to choose
from, what are you, the average user,
supposed to do? Well, today I'm going to
help you answer that question as I pit
Opus 4.7, GPT 5.5, and DeepSeek V4
against one another so you can see which
one actually makes sense for you. Now,
before we kick off this head-to-head
test between GPT 5.5 inside of Codeex,
Deepseek V4 inside of Open Code, and
Opus 4.7 inside of Cloud Code. Let's
first take a quick look at the
benchmarks, especially these two latest
models that dropped in the last 24
hours. Now, let's first talk about cost.
Now, DeepSeek V4, as you know, is an
open- source openweight model, but that
does not mean you can run this on your
computer because this thing is huge. I'm
talking 1.6 trillion parameters. You
need some serious hardware to run this.
So, we still got to pay for it. We're
still going to have to use the API, but
it is infinitely cheaper than the
competition, about 8 times cheaper. And
of the three models, the brand new GPT
5.5 is actually the most expensive,
which is kind of surprising because by
and large, OpenAI has been cheaper than
its anthropic competition. In terms of
what it will cost you per 1 million
tokens of output for GPT 5.5, it's going
to be $30. For Enthropic, it's going to
be $25. And for DeepS, it's going to be
$348.
Now, if we're talking about input
tokens, which is a smaller part of the
hole, GPT 5.5 and Opus 5.7 are the same,
it's going to be $5 per 1 million input.
And for DeepSeek, it's about like $1.70.
So, way cheaper on the input and way
cheaper on the output. That being said,
when it comes to 5.5, this is like twice
as expensive as 5.4. However, OpenAI
claims that it actually uses way less
tokens due to its power. So, while it's
double the price of 5.4, they say in
terms of actual token spend and actual
cost for the same task, it ends up only
being like 20% more expensive when it's
all said and done. So, just have that in
the back of your mind. So, we've talked
about the cost. Now, let's talk about
the benchmarks. How good are these
models on paper? I know we're all kind
of numb to benchmarks in general. we
need to take them with a grain of salt,
but it's still worth taking a look,
especially when we're looking at the
numbers that are reported by each player
on the same benchmark. So, there were
three in the coding category that all
three reported numbers that was Swebench
Verified, SweetBench Pro, and Terminal
Bench 2.0. Now, for SWEBench verified
and SweetBench Pro, Opus was the winner
there. On Terminal Bench 2.0, GPT was
the winner by far at 87.2, 2, which by
the way is a higher number than what
Anthropic reported for Mythos, Mythos,
sorry, which is kind of crazy. You know,
the super secret model they can't
release apparently does worse on
Terminal Bench 2 than GPT 5.5. Now, the
Terminal Bench 2.0 is the biggest
outlier here. Opus 4.7 and V4 Pro are
way behind, but take a look at Opus 4.7
versus V4 Pro. It's less than two points
while being eight times cheaper. And you
see the same sort of story here with
SweetBench Verified and SweetBench Pro.
Yeah, Opus wins, but when we compare the
second place with the third place, and
V4 is always third place, there isn't
the huge gap you would expect. I mean,
five points isn't nothing, you know, on
Sweetbench verified 85 to 86, but again,
eight times cheaper, open- source, you
know, there there's some actual
trade-offs here that we can make if we
don't need the most power. Another thing
that's interesting to talk about is long
context where oddly opus 4.7 is really
bad by the numbers like significantly
worse than 4.6 which kind of blows my
mind. And when we're talking about long
context where we're trying to retrieve
things between 500,000 tokens and 1
million tokens 4.7 is actually terrible
and does way worse than deepseek and GPT
5.5. Now you can have a whole discussion
about why are you even in the 500,000 to
1 million token range to begin with? how
many people are actually operating there
because we are hitting context rot no
matter what at that place no matter what
model you're using but it is interesting
that for whatever reason we've seen some
regression when it comes to the
anthropic models but big picture I think
the takeaway is 5.5 is really strong it
beats opus 4.7 in certain metrics loses
in certain metrics but it's an extremely
robust model and on top of that while V4
Pro is kind of you know lagging behind
by and large it's within striking
distance while being infinitely cheaper,
which again is a great option for your
average customer because right now it
feels like you don't have a lot of
options on the open source side that
actually can compete. Now, let's jump
into the actual head-to-head test with
all three of these models. And we're
using a harness for each of these
models. With 5.5, it's going to be
codeex. With Opus 4.7, it's going to be
Claw Code. And with DeepSeek V4 Pro, I
am using Open Code. And for the first
test, what we're going to do is we're
going to have them create a flight
simulator for us in 3JS that runs in the
browser. You can see the prompt right
here. I'm saying I want it to feel good
to fly. I want it to have some weight to
it. I want some strong visuals, and I
want it to use whatever structure and
tooling it thinks is correct. So, it's
straightforward enough that they know
what to do, yet there's enough leeway so
we can see some divergence between the
models. And while we are going to look
at what they're able to oneshot, we are
going to go through multiple iterations
of this and have follow-on prompts
because as cool as it is to see how well
it does on one shot, that isn't how we
really work in real life, is it? Right.
I want to see how it does when I give it
follow-on prompts and how quickly it
takes to get it to something I like. And
when we compare these three models,
there's really four things I'm going to
look at. It's going to be time. How long
does it take to build this? Cost, how
many tokens are we using? Quality, how
good is it? And then four is sort of
vibes, and that sort of relates to
quality. It's very subjective. Which one
do I actually like more? And also of
note, all three models, all three
harnesses are also using the exact same
skills. So, let's begin with Deepseek in
the questions. It's asking us, it's
asking what sort of flight model we
want. Let's go with full sim. It's
recommending oceans and islands for the
terrain. We'll go with that. Let's see
how. And then it's asking camera
preference. Let's do both. Let's see if
it's able to give us a toggle for both
first person and third person. We'll go
with its recommended tooling preference.
And we'll just go with a low poly model
for the aircraft and visuals itself.
Now, moving over to Codeex. Same sort of
questions, although it's only asking us
three. Saying what kind of flight should
this plane optimize for? Let's go with
hard simulation. Which playable
experience matters most for the browser?
Let's do island takeoff loop. It is kind
of interesting how they all have the
same one. And what camera and aircraft
presentation? I'm going to do toggle for
this as well. And for cloud code, we'll
do study sim learning. For the field,
ocean, and islands input, we will do
keyboard and mouse. It will let it go to
work. So plan mode by the large very
similar across all three. pretty much
the same questions of like what do you
want the physics to be? What do you want
the terrain to be? What do you want the
camera angle to be? So, no huge
difference there. And let's see what
they come back with in terms of a plan.
All right, so all three plans are
complete. So, let's go through each of
them pretty quickly and see some of the
differences. First one we're looking at
here is Deepseek and it's pretty bare
bones in terms of the plan it lays out.
So, gives us the project structure and
then talks very quickly about light
physics, environment, camera, and HUD
overlay and really just a few bullet
points. On the other hand, when we're
looking at 5.5 inside of codeex, it
gives us a summary, key changes, goes
into implementation details, the test
plan, and as well as the assumptions. It
spells all that out for us. And then we
have Claude codes plan, which took the
longest, took it about 5 minutes, but by
far is the most thorough because it's
the context, the stack layout, talks
about the flight model. It's going into
like the actual different moments,
talking about stalls, like the stall
buzzer, like it's it's going very very
detailed. goes into the controls, the
world, the mod, the actual aircraft
we're going to be using, performance,
and just keeps going on and on. So, very
detailed. So, now we're going to have
all three implement their plan, and
we'll see what the final result looks
like. So, GPT 5.5 inside of Codex was
the first to finish. So, let's see what
it looks like. So, here's the flight
simulator it got us. We have, you know,
some clouds in the sky. We have what
looks like an AOA indicator up there. We
have our speed down below. And let's see
if we can actually get this thing off
the ground. I will note there's no more
like runway. It's just like straight
grass and it said it was going to be
like an island thing. Although when the
camera kind of spazzes out, you can see
the runway down below there for a
second. All right, we're stalling out
and we just we can't even get off the
ground. All right, so this this one's
actually just a little is actually kind
of difficult. Um, so what I'm going to
do is I'm going to give it a second
prompt asking it to make it a little bit
easier to fly because it has a lot going
on here, but this is this is tough. So I
wrote, "It is really hard to fly. Can we
make this easier to use?" Aka a little
bit more arcadey and also the graphics
could use some work. So let's see how
that does. Now, of note, um it took 5.5
about 7 minutes to create that first
pass for us, and it took 63,000 tokens.
All right, it said it made it a little
bit easier to fly and updated the
graphics. So, let's see what the second
pass looks like. So, here's what we got.
Graphics definitely look better, but
let's see if we can actually get off the
runway this time. So, all right,
throttle's at 100%.
50 60. What's the rotation speed on a
Cessna? All right. 70 80 90. We got to
be able to get off the ground now. Okay.
Wrong way. Let's go. Get off the ground.
Get off the ground. Nope. This is
probably going to stall me out, isn't
it? Yeah. Stall. Okay.
This still needs this still needs some
work. So, let's let's give Codeex one
more shot. Let's give 5.5 one more
chance to make this actually playable.
So, I told it I can't even get the
aircraft off the ground and enter
flight. We definitely need to make it
easier to take off and actually fly the
thing. Okay, so it says it fixed the
takeoff problem. Apparently the brakes
started locked on before. I don't know
if that's why we weren't able to do it.
Oh, we it didn't automatically set it to
takeoff flaps. Yeah, this was we had
this on like super super simulator mode.
Here is attempt number three at our
flight simulator. Let's see how we do.
So, can we get off the ground? Oh, we're
bouncing on the runway with this time.
That's something. All right, cool. We're
off the ground. We're actually moving.
Let's see if we can get on one of these
rings. I mean, the graphics aren't that
bad, you know, for something just
generated in less than 10 minutes. Um,
it seems to be pretty accurate in terms
of, you know, it's giving me like my
vertical, you know, feet per minute down
at the bottom, my actual altitude, the
knots, heading, AGL. So, like it's
relatively sophisticated in terms of
tracking everything. I mean, this little
indicator in the front, I mean, looks to
be like an angle of attack, you know,
indicator, which is kind of cool. So, it
has some good stuff going on. The the
actual like controls are a little janky.
As you can see, I can't control this for
anything, but by and large, not bad. You
know, we can kind of like kamicazi this
and see what happens at, you know,
18,000 ft per minute.
But yeah, you know, for 66,000 tokens,
about 10 minutes, 15 minutes or so, give
or take, you know, with the back and
forth. I don't think that's bad at all.
So, now let's take a look at DeepSk. It
took about 10 minutes to do this. And in
terms of tokens, 63,000
44. So, 44 cents, 10 minutes. And here
is what Deep Seek came up with for us. I
have no idea
what I'm looking at.
This is supposed to be third person.
This is supposed to be the cockpit. And
obviously,
our first pass with Deepseek was an
utter disaster. So, I'm telling Deepseek
the simulator is a complete mess. The
graphics are completely buggy and I
cannot fly anything. Please fix. And
here's what our second pass looks like.
I still have no idea. Absolutely no clue
what the heck Deep Seek is. Oh, hey,
there's a plane. You know, there's
something.
I Yeah, this is This is brutal. And to
be honest, I feel like even giving it
another prompt to do this, I would need
to start getting very, very specific
about what we're trying to do, which
again like falls pretty short of what we
did with Codex. Like it was very, you
know, kind of bland prompts. was able to
get something at least close even on the
first pass. Like this clearly it's
completely struggling with the graphics.
We are just I I don't even know how to
describe this. But hey, it was super
cheap. So now let's take a look at what
Claude Code was able to give us. For
reference, it took 13 minutes to
actually execute the plan. The plan
itself took 5 minutes. So let's call it
20 minutes to come up with the first
pass. And then for total tokens, this
run took about 15% plus the 5% before
for plan. So we're looking at well sorry
we are looking at 11% context plus 5%
before. So call it 20 minutes 150,000
tokens for Claude Code which is
definitely the most expensive and
slowest out of all of them. And here is
Claude Code's attempt at this. Um for
whatever reason we are instantly in the
air. We are stalling. We are an IFR. I
don't know what's happening. We are
about to crash something. Let's Can we
save this? Can we pull this out of a
dive? No, we're stalling. No, we're
dead. Okay, that's interesting. Um,
again, it instantly slingshots us into
the air. We are in the clouds. We are
stalling. I don't know what is
happening. We need We need a second
pass. So, I wrote, "Upon loading, I'm
instantly thrown into the air. It's hard
to control. I want to start on the
runway, and I want it easier to fly. Oh,
and by the way, improve those graphics,
too. So, it took about four minutes, but
it made some changes. We're going to
spawn on the runway. It changed the
gear. So, now it's tricycle gear and a
few other stuff. So, let's see what it
looks like. All right, so here it is
again. We are thrown immediately into a
fog bank. I'm trying to control this
thing. And I just Yeah, there's there's
no controlling this at all. All right,
we are going to give we're going to give
Cloud Code one more chance here. So, I
told it's still instantly slingshotting
me into the sky. I said, let's go with a
much more arcade type feel with the
controls. I think we probably should
have done that with the initial prompt
for all three. I think going for a more
realistic sim type thing, it it really
struggles to
I think do that in a way where it's
still user friendly. I think it's
probably doing a good job under the hood
in terms like okay like angle of attack.
All right, you're stalling at this you
know angle versus the speed and all that
but actually manipulating this from the
computer is basically impossible.
Although I I think the fog stuff is
really strange. So, let's see if after
the second round of prompts, it's able
to do a little bit better because right
now GPT 5.5 did much much better. So,
Cloud Code made some more changes, made
it more user friendly, and let's see if
I'm still going for my instrument rating
this time. So, yep, we're still going
we're still going for instrument rating.
RIP men's here, but you know, I can kind
of see it. You know, I can I can check
my instrument panel. All right, we're
coming off the runway.
Um,
yeah. Okay. Can I Why is there a tree in
the runway? Trying to trying to go up.
Can I go up? Can I pitch? Click canvas
to lock mouse. I What? Oh, we're in the
We're in the air.
Nope. Nope. We died. So, yeah, I think
this one is pretty clear. Uh, GPT 5.5
easily the winner. I think Cloud Code
was second place. I would give it second
place. You know, it definitely struggled
even with the prompts we gave it. We
didn't give it great prompts. Let's be
totally honest. I think given more time,
better prompts, a few more back and
forth, we could have got it to where we
wanted to go. Like, it was at least it
had an aircraft, it had a runway, it had
trees in the runway, but it had the the
actual things we needed versus Deepseek
with Open Go. I have no idea what was
going on there. That was a complete
mess. I feel like I would have had to
start over from the beginning, like give
it a very specific prompt. like it
wasn't even close to being messed with,
but GPT 5.5 right off the rip, you know,
with pretty vague prompts. I thought it
did really good. 5.5 also used a total
of 66k tokens. We're looking at over
here with Opus altogether about 200,000
tokens. So, a quarter of the tokens
essentially quarter the cost and it was
a bit faster. I mean, at this point, I
don't even care about how open code
actually took longer than GPT 5.5 as
well. And it just sucked. Let's just be
honest, it just sucked. Now let's move
on to test number two. This time we are
going to be asking them to create a
landing page that shows off web GPU
shader work using 3JS. Now web GPU
shader work is the kind of stuff you see
on awards websites. I'm talking websites
like Igloo. This kind of thing like very
high-end graphics. It looks like a video
game. It's essentially using your
computer's graphics card to render all
this stuff. Now, I don't expect any of
these to get anything even close to what
we see here, but I want to see what they
can do using essentially the shaders
technology. This is definitely a step
above your basic SAS templated landing
page. I want to see what they can do and
push them to the limits in the world of
web design. Now, I've given all of them
a skill that actually breaks down how to
do this sort of thing. So, it's not like
they're completely in the dark and one
also doesn't have an advantage over the
other. The only thing I've told them is
I want it to feel modern and visually
striking. something you would see on
awards and to make smart use of GPU
compute. So they can pick whatever stack
and project structure they like and use
good judgment on hero concept UI and
interactions. And just like the first
test, they're all on plan mode. So let's
get started. Okay, so they all finished
their plan and funny enough, none of
them asked me any questions even though
we put them in plan mode. So let's take
a look at GPT 5.5 first. So it's telling
us it's going to do a fullbleleed
interactive GPU driven hero. The concept
will be a living signal field with some
like dense particle thing it's going to
do. We'll see what that ends up looking
like. And overall, it's a minimal awards
style landing copy. Fully interactive
web GPU scene with pointer reactive
compute simulation.
All right, for Deep Seek, it's a pretty
short and sweet plan just like we saw
with the flight simulator. Hopefully, we
get a better output this time. But a
hero section with 75,000 GPU computer
particles. I am kind of guessing that
all of them are going to go for some
sort of like particle theme on the on
the hero.
So, it's gonna have mouse interaction
integration. It'll have a one-time
initialization. And then we should see
stuff like bloom, chromatic aberration,
a custom vignette, and some film grain.
So, we'll see what that actually ends up
looking like. And then we have Opus
4.7's playing again going for this
particle thing with bloom, and it's
going to be interactive with the mouse.
So, we'll see if any of these actually
look different because on the surface
all their plans sound very similar. So,
the first one done was 5.5. It took
about 6 minutes. And in terms of tokens,
we've used 107K. So, let's see what it
built us. And here's what it created for
us. Now, this is very bright. Um, so
it's hard to even see the actual
particles, but you know, as we scroll up
and down, it does have an animation
going on in the background as well as,
you know, some subtle color changes.
Looks like right now our mouse is
supposed to attract the particles and we
have I'll move this over here. It gave
some options for like repelling it
versus drift, but again, it's kind of
tough to to see it um due to how bright
it is. So, I told that it's hard to
actually see the particles due to the
brightness. It also takes over a lot of
the hero text. So, can we turn down the
brightness a bit and also push it to the
right a bit more because right now it is
kind of overpowering. You can't even
really read the text over here on the
left due to just how freaking bright
these particles are. And here's the
update after the second run. It's a
little bit better. It isn't as
overpowering and leaves some room for
the text. Um, although I will say it
it's kind of blurry almost, but you
know, it's not bad. Like it it set out
to do what we told it to do given the
somewhat vague prom. So, I'm not blown
away by sort of the design it came up
with, but I'm not like upset about it.
Now, let's take a look at Clawude Code
because as we've been doing all this,
Deep Seek is still over here in the
trenches trying to figure this out. And
here's what Claude Code gave us.
So,
kind of nothing.
I'm not sure if it's saying the
background. I guess the entire
background is supposed to be
the WebGL. I'm I'm assuming it's very
understated,
which I guess is something you could
totally do. I mean, like on screen it
doesn't look like it looks kind of cool,
but I I'll be honest, I was looking for
something a little more flashy. So on
the second pass when I told it to make
it a bit more flashy, there wasn't a
huge difference. Although like it's it's
really subtle. There's kind of like this
film grain almost like this blur that
goes from bottom to top. So it's a
pretty subtle thing. And you can see
here on the bottom it tracks like the
frames per second. It's using 250,000
particles. So I mean honestly it looks
cool. It's just not super flashy. So
it's definitely like a taste thing. Now,
total tokens on the cloud code side was
about 175,000 and it took just slightly
longer than 5.5 inside of Codeex. Now,
let's take a look at DeepSk, which has
taken 116,000 tokens at this point. It
took the longest um as well, but total
cost, we're talking again under a
dollar. And here's what it gave us. So,
it's
kind of this particle field thing that
uh somewhat follows my mouse.
Interesting. I think it might give you
like an epileptic seizure,
honestly. Um, beyond that, it's pretty
bland. Um, the flux, you know, text
right here kind of changes colors, but
yeah, pretty much just created this this
thing. And after telling Deep Seek to do
another pass, it then came back with
this where now it kind of has like some
weird parallax thing. It's got some like
blue stuff going on in the background.
And now this thing looks like a UFO,
which kind of responds to your mouse,
but
yeah, it's it's it's something. And
overall, the token count from Deep Seek
was 130K tokens, coming in at a $143.
So, after all those tests, where does
that really leave us? So, now let's talk
about the final results. When it comes
to test number one, which was the flight
simulator, clear winner. That was GPT
5.5 inside of Codeex. It was quicker
than Opus 4.7 inside of Cloud Code. It
was also faster and the end result was
by far the best. Deepseek did terribly
in the flight simulator. It wasn't even
close to what we were trying to do. I
would have had to continue to prompt it,
prompt it, prompt it to even get it to
like close to the first pass from 5.5.
And Opus 4.7 and Cloud Code was like,
eh,
it wasn't awful. like it really didn't
work at the beginning, but after a
couple prompts, you could tell we could
get it to a place where it was
equivalent to what GPT 5.5 was doing,
but that would have taken more prompts.
It would have taken more time, and
ultimately would be more expensive. So,
clear winner for 5.5. In terms of the
web GPU landing page, again, DeepSeek
struggled here. I was not a fan of this.
I don't really know what this is
supposed to be. Sure, I didn't give it a
super great prompt, but like is this
what we're going to be getting as a
baseline median outcome if I don't like
grab Deep Seek by the reins and really
force it to do something? I guess so.
Now, when we compare Opus in 5.5, I
would have gone with Opus 4.7 and Claude
Code with how it handled the web GPU
thing. I think that has to do with sort
of a taste kind of deal. Yeah, you could
argue the 5.5 was flashier, but I
thought it was kind of ugly. Um, again,
in all these tests, we kept the prompts
rather vague to see what sort of path it
would go down. So, I would definitely
give Opus the lead here, although it was
more expensive and it also took slightly
longer. So, if they were given a more,
you know, hands-on prompt that was very
specific about what she wanted to do
because 5.5 did what we wanted it to do,
like it did create a web GPU landing
page. I just thought it was ugly. So, it
still completed the task. It just didn't
complete it as well, I think, as Opus.
Now, big picture, what does it mean if
we take all that together? Well, I think
it means great news for anybody who's
using agent coders. We have options,
right? You can use Opus and Claude Code
or you can use GPT 5.5 and codecs.
You're not wrong with either. I think
it's totally a personal preference at
this point. And the best part is if you
go down the cloud code route, it pretty
much all applies to Codex. If you go
down the codeex route, it pretty much
all applies to clawed code. So, I don't
really think there's vendor lock in the
sense like, oh, I've only learned about
clawed code, like I can't go to codeex
or vice versa. That's not the case at
all. If you're doing this the right way,
what you're really learning is AI
fundamentals and how to build things.
And that applies to both of these guys.
And the more competition, the better it
is for us, the consumer. Now, as for
DeepSeek, n I don't know. I wasn't very
impressed. This might be a situation
we're like, okay, like Deep Seek makes
sense if we're doing simpler tasks where
we just don't need the power of
something like Opus or we just don't
need the power of something like GPT 5.5
because remember we're talking about
something that is eight times cheaper.
Sure, I didn't like the web GPU landing
page this thing came up with, but was it
eight times worse?
Maybe, maybe not. Kind of hard to
actually, you know, articulate that and
quantify that, but obviously that's
something we need to take into account.
So, you know, I don't think it's really
competition to be frank with 4.7 or 5.5.
I think though if you're doing simpler
task and you're like very token
conscious, very cash conscious, then
hey, maybe Deep Seek makes sense for
you. So, that's all I got for you guys
today. I hope that sheds some light on
these three models and how they kind of
stack up to one another. I think it's a
great time to be in the space. More
competition is better for everyone. So,
as always, if you want to get your hands
on the Claude Code Masterass, make sure
to check out Chase AF Plus. There's a
link to that in the description and I'll
see you