YouTube Transcript:
Master Any AI in 2025: Ultimate Comparison Guide (from ChatGPT to Gemini)

Large language models are rewriting the
playbook. And if you're not riding the
wave, you're stuck on the shore. But
who's time to test drive every single
model out there. That's why we've done
the heavy lifting for you. This video
will speedrun the hottest LMS, break
down what they're best at, show you
exactly how to prompt them, and drop
insider tips you won't hear anywhere
else. By the end, you will know which
model fits your style, and how to make
it sing. This is the last LM crash
course you'll ever need.
Chad GBT is the goat. Everyone knows
that. Take any person's laptop and you
will have Chad GBT open in a separate
tab. It was the first one to hit the
market and all other LMS were modeled
after it. The interface, the features,
the internal logic. All other LLMs get
new features only after Chad GBT gets
them. And in this channel, we have tons
of videos already about everything you
might want to know about Chad GBT. And I
suggest you watch those. Here's our full
guide to the GBT 4. It's a halfhour
crash course into chat GBT. Here's
another one where we go over all the new
features like canvas task scheduling or
deep research. And here's our latest
guide to the new image generation in the
GBT40. So if you want to learn CHBT,
watch those videos. And for learning
generative AI in general, welcome to our
brand new course into generative
AI. I know you might imagine typing over
your webcam and mic whenever you hear
meta, but trust me, it's actually pretty
cool. Meta AI comes in two versions.
There is a separate website and there's
an AI that's built right into Meta's
larger platform. The web version is
pretty simple and has almost everything
you need. I'd put in about the same
level as Sorro when it comes to replies
and features. Of course, it can write
articles for you or blog posts and it's
right in style is pretty close to Chad
GBTS. Meta AI runs in a model called
Lama and they've been updated like
crazy. bigger context windows, improved
ways to handle different kinds of
content, and better logic. It's honestly
a good model, and the fact that they let
people use it for free is a nice move.
Feature-wise, Meta AI in the web is
really flexible. Can handle nearly
anything you throw at it, like images,
PDFs, text documents, and data files. It
also has a memory you can update right
in the chat. I haven't seen a special
menu for managing or wiping that memory
like Chad GBT has, but it still
remembers whatever you tell it. It
doesn't pull info from different chats
though, so I'd recommend using that one
conversation as your go-to place for
instructions. Just start with remember
that and then list your preferences.
There is a canvas mode too which feels a
bit like Chad GBTS, but it's more like a
write in workshop. You can highlight any
bit of text and ask for a rewrite,
change the entire text or drop in
images. It's pretty fun. Just highlight
a word, tap imagine, and the AI
generates four images right away for
that spot in the text. You can even
tweak the prompt to change what it
shows. The canvas editing window itself
is bigger than chat GBTS with a smaller
chat window. It's missing some
convenient controls, but it does have
formatting tools. Overall, it's a nice
feature, especially if you like to
write. And of course, if you're aiming
to replace chat GBT, you need image
generation. Meta AI has that covered.
It's sort of like Adobe Firefly. You can
generate an image in the chat, pick the
one you like, and then edit it if you
want. You can also adjust the settings
for all four images, like changing the
aspect ratio, lighting, or style, and
you'll see them update in real time.
It's not the best image generator out
there, but it's solid and gets the job
done. It struggles a bit with putting
text and images, though. That's the web
version of Meta AI, but Meta AI has
really grown into an entire ecosystem.
You can even use it in Facebook chats
and comments or through Meta's glasses.
I'm not the biggest fan of everything
Zuck does, but you have to admit it's
neat. Can find info, handle vision
tasks, all that good stuff. The glasses
are basically the full Meta AI
experience, but you don't actually need
them to use the tool. Even without the
fancy hardware, Meta AI is a strong
option. AI is the mind behind all the
innovations. If you want to master these
skills and understand how these tools
are built, or maybe build one, then
Skillup is the platform to learn these
skills. They're sponsoring this video
and Skillup by Simply Learn honestly
feels like it was designed for all of us
AI people. Whether you are a student
who's just dipping your toes into the
world of tech developer trying to level
up with the latest tools or even a
manager looking to stay ahead of the
curve, Skillup has you covered. They've
got a massive library of self-paced
courses and everything you can think of.
AI, data science, cyber security, cloud
computing, digital marketing, project
management, you name it. And these
aren't just slapped together lectures.
Each course is built in collaboration
with industry pros and top tier partners
like Google, Microsoft, and AWS. So,
you're not only learning the theory
behind these topics, you're getting the
kind of knowledge that's actually used
out in the real world. One thing I
absolutely love about Skillup is how
flexible it is. I mean, I've been
watching videos on my iPad while waiting
for a coffee to brew and even sneaking
in a few lessons on my phone. It's all
self-paced, so you don't need to worry
about juggling strict schedules. And
here's the cherry on top. Once you
finish a course, you get a free
certificate to show off your new skills.
No hidden fees, no annoying strings
attached. So, if you're looking to pivot
your career into tech, boost your skill
set in your current job, or maybe just
learn something totally new for fun,
Skill Up is kind of a no-brainer. You
can sign up using the link in the
description below or the pinned comment.
And when you do, let me know which
course you pick.
Answer this. What is a Chad GBC killer?
Is it just some tool that does
everything better? I think can also be a
tool that lets you do a bunch of stuff
in one place, like using different AI
models under one subscription. That's
exactly what PO is. And it's really
interesting if you know how AI tools
differ and which ones specialize in
what. With PO, you can switch between
them to get better results. That's the
whole point. You're not stuck with the
single model. You basically get them
all. The list of models is crazy. I
scroll and scroll and it keeps going.
Lamok, Grock, Gemini, Deepseek. I can
even use Chad GBT 3.5 here. Then each
model has its own chat with most of the
features you'd expect. Multimodality is
on board. So I can upload images for
analysis, PDFs to grab text from, and so
on. Which features you get depends on
the model. So if it can do something, Po
won't magically force it. But all the
usual prompt and tricks still work like
they normally do for each model. It's a
great way to explore AI and figure out
which model fits your needs best. I love
how flexible PO is. If a model has
canvas, PO will open it without any
trouble. If a model like 11 Labs can do
audio, gives me simple audio tools. If
the model needs more reasoning, it does
that too. I can even create my own apps
because of the built-in cloud
integration. For instance, I made a
quick app for removing the background
from an image. One click and I'm already
uploading my photo. PO has a model of
its own, but it's just okay.
Fortunately, each response has a button
underneath to compare that same response
with another model. So, unlike with Chad
GBT, you can swap to a better model in
the middle of the chat if you feel like
it. Don't say you weren't waiting for
this one. Gemini 2.5 Pro. For years,
Gemini has been a mustave in videos like
this, but this time it really deserves
its spot. Right now, it's pretty much
what GBT5 could be, except it's from
Google and free to use. You get up to 50
messages per day with its top model.
Then it drops to smaller one that's
still almost as good as 40. Until GPT
4.1 arrived, the simplest advantage of
Gemini was its 1 million token context
window. Now, Chad GBT has cut up to that
number and Google is only weeks away
from doubling Gemini's token limit. By
the time you're watching, this might
have already happened. Gemini can
remember a ton of information at once,
which is a big deal if you like long AI
chats, maybe you're planning project,
summarizing huge sections of text, or
building each step as you go. Chat GBT
does have memory, too, but it might lose
some details if your conversation
stretches on too long. Gemini can handle
more data without forgetting, which
helps a lot when you're writing lengthy
essays, digging through piles of
information, or needing the AI to recall
earlier points from your discussion.
When it comes to searching online,
Gemini often goes a step beyond Chad GBT
because web search is on by default and
also works with its best model. It's not
just about giving a quick answer. Gemini
can pull info from different places and
mix it into one detailed response even
if you didn't specifically ask. Sure,
there is a deep research mode as well
and it performs as well as Chad GBTS,
but most of the time you can get what
you need without it. One of the best
things about Gemini is how smoothly it
ties into Google services. You can
connect it to Docs, Sheets, Drive, Maps,
and plenty more. That means you can grab
notes from Drive, ask for directions, or
gather travel tips without juggling
multiple tabs. Chad GBT can do something
like that with plugins, but those can
feel like an extra layer that doesn't
always work perfectly. The only area
where Gemini lags behind Chad GBT is
image generation. Don't get me wrong,
it's solid here, just not as good as
GBT40. Sometimes it messes up text,
messes a part of your prompt, or isn't
totally consistent. Still, for a free
image generator, it's quite good. And to
your surprise, Gemini is actually better
than Chad GBT when it comes to
multimodality. It works with a wider
range of files. Of course, Gemini
handles large PDFs, word docs, or text
logs easily. You can attach these files,
mention them in your question, and
Gemini will summarize or analyze them.
It also works with images, audio, and
even videos. Yes, videos. Its coding
mode is also stronger. It's fantastic at
debugging. supports popular coding
languages and can even run code right in
the tool. Look at this game it fixed. No
plugins needed. Just paste the code and
it works. And sure, Canvas supports code
in here, too. Gemini today isn't the
same Gemini we knew year ago. Now, it's
leading the pack and OpenAI will need
something really special for GBT to
catch up to
Google. Grock was first released a few
years ago and has changed a lot since
then. Now, it's a worthy alternative to
Chad GBT, especially considering how
much you get for the price compared to
the features. For starters, it has
pretty much all the same essential
abilities as Chad GBT, can write
articles and do web searches just as
well. So, that alone isn't the big draw
anymore. Rock works a lot like Chad GBT
and serves as a perfect example of
generative AI, specifically an LM. Since
most LM share the same basic principles,
learning the fundamentals of generative
AI will let you use almost any LM
effectively. That's exactly what we
focus on in our brand new 101 crash
course into generative AI at Geek
Academy, where we show you how AI
interprets prompts, how to prompt
properly, and what common mistakes to
avoid. We're adding new lessons every
week and the course covers everything
from the inner workings of AI and the
logic behind tools like Chad GBT to
in-depth tips and prompting for image
generators complete with concrete
examples and insights into their popular
features. We also explore developer
tools and coding assistance in real life
situations plus essential prompts and
templates. Beyond that, we dive into
music and video generation tools, AI
avatars, texttospech options, and so
many others. Basically, if it has to do
with generative AI, it's in our course
guiding you from zero to pro under one
Geek Academy subscription. And right
now, we're offering a massive 80%
discount on a 6-month access to Geek
Academy. It's a limited time offer, so
don't miss out. What really stood out to
me was how convenient and logical Grock
feels. If I want to research something,
I have three ways to do it. One is to
just ask a question normally, which
triggers a quick simple web search. The
second option is deep search, which
takes around a minute and pulls together
more thorough and concise info. Grog
gathers data from various articles and
in other half a minute puts together a
solid chunk of text with conclusions and
summaries. The last option is deeper
search, which is basically Grock's
version of deep research in Chad GBT,
but way faster. The request I tried
would have taken Chad GBT at least 5 to
10 minutes and Grock did it in three and
a half complete with links, detailed
info, and neat formatting. Another cool
feature is that I can turn on reasoning
anytime without having to switch models.
This reasoning mode works about as well
as GBT1 or GBT3, only faster, and it
doesn't cost anything extra. By now,
Grock is basically that omni model. GBT5
is aiming to be one model that can do it
all. It writes, handles files, does web
searches, and even generates images all
from the same place. Image generation
here is pretty cool, too. I can upload a
picture and make edits like adding
glasses or upscaling. And unlike the
latest Chad GBT image tools, Gro keeps
everything consistent. Chad GBT
sometimes shifts the whole image around
when you edit it. Granted, Grog might
not be as sharp as Chad GBT at
generating text and images, but the
overall edits work really well. Creating
new images is super simple and follows
the same rules as Chad GBT. The editing
window just works differently. Grock
doesn't let you pick a specific part of
the image to edit. Instead, you choose
between subject, background, or style.
Once you pick, you don't confirm
anything. The changes appear almost
instantly. There is also a prompt for
bigger tweaks. I really like this image
generation. It's consistent. It's
reliable and free. I agree that Grog has
fewer flashy features than Chad GBT, but
some parts are just as good or better.
Consider the workplaces for instance.
They're basically the same idea as
spaces in Chad GBT. You have your own
files, your own chats, and your own
custom instructions for every
conversation in that workspace. There
aren't a ton of settings to tinker with,
but you can set your own custom
instructions or just presets for
different response styles. You can also
manually switch gro into one of the
suggested roles like specialist, doctor,
or therapist. It's basically the same as
typing a prompt beforehand. It's still a
handy extra. I really like Grog. I
definitely need to do a full video on
it. Oh, wait. I'm already working on
that. So, subscribe if you don't want to
miss it.
I always used to wonder why can't I run
Chad GBT right on my own laptop. I know
the model is huge and needs tons of
resources, but come on, wouldn't that be
cool? Well, guess what? Now I've got
Chad GBT on my MacBook through a
console. The only catch is that it's not
really Chad GBT at all. It's a tool
called Deepseek. The biggest advantage
is that Deep Seek is totally free. Chad
GBT hides his best stuff like advanced
reasoning and unlimited image generation
behind a monthly fee. Deepseek doesn't
charge anything for advanced reasoning
or any other main feature. You sign up
and you get everything and offers no
payw wall. Right now the tool does lack
some of Chad GBT's extras like image
generation, deep research, and canvas,
but it does have advanced reasoning and
web search. Even though the web search
doesn't always work perfectly, that
reasoning feature is basically a direct
clone of GBT1, but it's actually better.
When you ask it something, it breaks
down the steps in a little thought
process panel that you can see. This
step-by-step method makes the AI's
answers clearer and more accurate. It
takes about as long as GB01 with results
just as thorough, clever, and original.
Then, DeepC goes further, can handle
files. GBT01 can work with images that
have text in them, but can tackle an
Excel data set. Deepseek is exactly what
you use if you want to parse data, find
patterns, or spot correlations, and
because it runs locally, it's a perfect
sidekick for data analysis. It can
generate graphs yet, but will probably
change soon. Those same offline
abilities also make it great for
developers. You can ask for a simple
HTML layout, a small Python script, or
even a basic JavaScript game. Deepseek
doesn't just generate the code, can
actually run certain demos right in the
chat. For example, it might whip up a
quick snake game and you can play then
and there. Think about how much easier
coding could be with a model like this
running in the background. Granted, you
would need a powerful, pretty powerful
computer, but it's still awesome. How do
you install it? All you need is a single
app, bit of storage, and a few clicks.
Then you're all set and your data stays
on your machine. You install it using a
tool called Alama which sets up DeepSeek
for you without a bunch of technical
headaches. Don't forget to check out our
full Deepseek guide if you want detailed
instructions on how to do it right.
Deepseek also has a mobile app. It's not
as slick as Chad GBT's official app, but
it does have the same features as
Deepseek on on your computer like
reasoning, data analysis, and web
search. It's laid out pretty much the
same with a simple chat window. Two
toggles for advanced reasoning and web
access, plus a button for adding files
or snapping pictures on the spot for
OCR. I've only found two real issues
with DeepSeek. One, the servers can be
busy a lot because there's so much hype.
It's not as bad now, but sometimes
you'll still see a server busy message.
And second, if you want image
generation, you've got to use another
tool like Janice from the same
developers. Janice doesn't give you a
ton of control, but it can generate high
quality images pretty fast with no fuss.
I really hope they bake that into
DeepSeek soon. I'm not fully switching
over yet, but I keep it on my Mac for
those moments when I'm working on with
private files or stuck without internet
access. I get this question from friends
all the time. I don't like switching
models in CGBT. What should I do? And my
answer is always the same. Use Claude.
The latest version does all the model
picking by itself. For easy stuff, it
uses simpler models. And for more
difficult tasks, flips to its reasoning
model. And by the way, that reasoning
model is right up there with GBT1, but
with 03 level speeds. Claude isn't just
CHBT clone, and it can really shine in
areas where Chad GBT might slip up. You
might notice something about Claude's
responses. is they tend to feel more
thoughtful, more careful, and sometimes
more detailed than Chad GBTs. This
reflective quality isn't just random.
It's because Claude is trying to weigh
context very carefully and lean toward
clarity over confusion. Claude's big
context windows might sound like a
technical detail, but they actually make
a big difference in keeping the AI
focused. It's better at sticking to
single narrative over long
conversations. I tried copying entire
chapters of text or big sets of data
into Claude, then asked specific
questions about each part. It rarely
mixed up questions or forgot what it
read and it really seemed like it was
actually holding on to older messages
instead of guessing once things got
complicated. If I want to compare it to
Chad GBT for writing, Claude can
definitely keep up. Its paragraphs
usually feel more structured and
cohesive. Chad GBT sometimes loops back
or jumps around if you throw it complex
prompts while Claude breaks ideas down
more directly. Both AIs can produce
usable text, but Claude's calm and
organized style can come off more
purposeful. That said, Claude could use
an upgrade in a few areas. It needs to
learn how to generate images and handle
them better, and its OCR could stand to
improve. It also needs better
multimodality. Right now, it doesn't
support a ton of formats and it and
isn't amazing at data analysis. As for
prompting, Claude is good at
understanding natural language, but some
of its prompt and practices feel
outdated, like something from 2020. We
have a whole guide on our channel about
Claude, plus some handy PDFs and posts
over at Geek Academy if you want to dive
deeper. Another interesting point is how
Claude deals with writing style
preferences. If you pace your own
writing and ask Claude to mimic it, it
usually does so without sounding forced.
Chbtt can do that, too. But sometimes it
goes too far or misses the subtle things
in your wording. Claude is better at
picking up on those little hints like
how fast or slow your sentences flow and
exact tone you're aiming for. That might
seem like a small edge, but if you need
the AI to match your personal voice for
a big writing project, it's really
helpful. Sometimes Claude will refuse to
answer certain questions, but that
doesn't happen much in everyday use. Of
course, Chajiti can still be more
playful or creative if you're just
messing with it for stories or
brainstorming. Claude can be
imaginative, but usually balances that
creativity with a bit more logic and
caution. Mestral is a French AI that's
been quietly improving behind the
scenes, and now it's finally a solid
option you might actually want to check
out. I won't pretend it's flawless. No
AI is. But what's great about Mistl is
that it kind of reminds me of Chad GBT
in in its early days before it turned
into the enormous beast it is now.
Mestral is simple, almost bare bones,
and that's part of its appeal. When it
comes to general writing, it's basically
on the same level as GBT40 in terms of
quality and depth, but Mistral is
faster. It splits out answers in just a
few seconds, much faster than Chad GBT
can do. Web searches, look over your
code, and generate images. It even has a
canvas feature. The catch is that using
Mistral can feel a bit awkward because
even if you switch on all these tools,
you still have to actually call them out
by name. Canvas doesn't pop up
automatically, so you have to literally
say use canvas. Still, it's a pretty
cool mode. You can highlight chunks of
text to rewrite and you get handy
controls for length, style, and other
editing settings. I like how the canvas
stays in the middle while the chat
window shifts to the right. Just don't
try generating images inside canvas. It
won't insert them the way Meta AI does.
As for coding, it's decent, but nothing
too advanced. It can look at code
catchbooks and point out errors, but it
doesn't have built-in frameworks to
preview your app. So you can't see a
live version of your code like you can
in Gemini. Personally, I wouldn't rely
on Mistral for coding projects. Image
generation is okay. Nothing
earthshattering. It follows prompts well
enough and the results look all right,
but it's it'sn't a lot of controls. I
also notice it's more sensitive to
detailed prompts than other image
generators. You really have to specify
style, framing, composition, that sort
of thing. So, no, Mistral isn't going to
replace GBT40's new image generator
anytime soon. One area does shine is
file handling. You can build a little
library for each chat, making it simpler
to refer back to those files later on.
That's a small but handy
feature. My favorite feature in Chad GBT
has always been deep research. I'm
serious. Every video we make starts with
deep research, but it costs money. So,
for a free option, I always suggest
Perplexity. It may not be a GBT killer,
but it's definitely a deep research
killer. Perplexity is really a research
tool at heart. It doesn't try to do
every single thing. It just does
research really well. Its results are
almost as accurate and in-depth as Chad
GBTS, and it's basically free. It
doesn't go for Chad GBT's friendly tone
either. Focuses on delivering factual
answers in a direct nononsense way. I
actually appreciate how perplexity gets
straight to the point. I agree that file
handling is better in Chad GBT. It's
fully multimodal and perplexity is more
limited. Perplexity isn't designed to
process huge data sets, but it's perfect
for PDFs, short code snippets, or
smaller word docs. You upload them,
describe what you need, and the AI
either summarizes or breaks the text
into manageable parts. Perplexity also
lets you group these files into spaces
which act like folders for your chat
threads, documents, and even
screenshots. Going back to research, the
basic mode is good, but Perplexity also
has a proarch mode which works like the
deep research we all know and love, but
instead of Chad GBT's methods, it uses
its own algorithms. So, you'll usually
see a different approach. If you turn it
on, it makes Perplexity dig deeper and
use more advanced reasoning. And instead
of hiding footnotes or skipping them
altogether, Perplexity shows its sources
right at the top of the reply. This is
one area where it might actually do a
better job than Chad GBT for those who
really want transparency. But there's
more. Focus is a neat feature that
tailor the system for different types of
data. There are five options: web,
academic, video, social, math, or
writing. In academic, for example,
Perplexity pays special attention to
peer-reviewed articles and well-known
research. Perplexity's coding skills are
practical, but they're not the
centerpiece. You can paste a Python
script or a C++ snippet, and it will
point out small bugs or suggest minor
improvements. A few parts of Perplexity
might feel niche, like the pages tool
that turns your chat into an
article-like format. You might not use
that every day, but it can be useful now
and then. I love Perplexity. But to
really make the most out of it, you
should stick to research tasks. For
everyday stuff, I'd still choose Chad
GBT. Over the years, Chad GBT has become
such a big name that it's hard to find
true replacements. Every new tool tries
to match it or outdo it. Some do manage
and some don't. GBT5 will probably be a
fantastic model and might pull ahead of
the pack for a while until Google or X
AI steps in again with an update. So,
would you really ditch Chad GBT for
something else, especially after you
learn how to use it well in our new AI
crash course? Thanks for watching and
see you in the next video.