Kind: captions
Language: en
You're probably tired of waiting months
for AI companies to release their next
big model only to be disappointed by
incremental updates that barely move the
needle. Well, here's the thing about
XAI's Gro. While everyone else is taking
their time, Elon Musk's team has been
releasing major updates every few
months.
And what I discovered about Grock 5 is
going to completely change how you think
about the AI race.
Welcome back to bitbiased.ai, AI, where
we do the research so you don't have to.
Join our community of AI enthusiasts.
Click the newsletter link in the
description for weekly analysis
delivered straight to your inbox.
Everything we know about Gro 5, Elon
Musk's next AI model that he claims will
be crushingly good and could potentially
achieve AGI.
We'll cover the insane timeline they're
working with, the rumored features that
make other AI models look outdated, and
why this release could be the most
significant one yet.
First up, let's talk about how XAI went
from zero to potentially beating
everyone else in just 2 years.
The Gro speedun. Here's something
that'll blow your mind. While most AI
companies take a year or more between
major releases, XAI has been operating
on what I can only call an insane
development schedule. Gro 1 launched in
November 2023. Basically XAI's entry
into the chatbot arena inspired by the
Hitchhiker's Guide to the Galaxy
till you see this timeline.
Just 6 months later, in May 2024, Grock
1.5 dropped with a massive 128,000 token
context window and significantly
stronger reasoning capabilities.
That alone would have been impressive.
But they didn't stop there. Only 3
months after that, in August 2024, Grock
2 arrived.
And this is where things got
interesting. Gro 2 didn't just improve
on the previous version. It completely
transformed the model.
We're talking major gains in coding,
reasoning, and vision capabilities. It
could generate images using a built-in
flux model. And here's the kicker. It
matched or even beat GPT4 Turbo on key
benchmarks.
Think about that for a second.
A company that had only existed for
about a year was already competing with
OpenAI's 2025 and XAI unveiled Gro 3.
This is where the compute power gets
absolutely ridiculous. They trained this
model using roughly 10 times the compute
of Gro 2, leveraging their 200,000 GPU
Colossus cluster.
And they introduced these new think
modes for advanced reasoning, basically
letting the model work through complex
problems step by step. Then came July
9th, 2025.
Gro 4 launched and XAI literally called
it the most intelligent model in the
world.
Now, I know every AI company says that,
but Gro 4 brought something genuinely
different to the table. Native tool use.
Not bolted on features, but a code
interpreter and web search functionality
built directly into the model's
training.
It also bumped the context window to
256,000 tokens in the API and added an
upgraded voice chat mode that can
literally see what you're looking at and
analyze live camera input.
Timeline absolutely bonkers.
Grock 1 to 1.5 took 6 months. Grock 1.5
to Grock 2 took only 3 months. Grock 2
to Grock 3 took 6 months. And Grock 3 to
Grock 4 took just five months.
They've been releasing major versions
roughly twice a year with each one
bringing qualitatively new capabilities
that completely change what the model
can do.
And here's where the story takes a wild
turn because Gro 5 is coming even
faster.
The Gro 5 timeline.
In early August 2025, Elon Musk dropped
a bomb on X. He posted that Gro 5 will
be out before the end of this year and
it will be crushingly good.
This wasn't just casual speculation. He
said this directly in response to
OpenAI's GPT5 announcement, essentially
throwing down the gauntlet. But that's
not even the most interesting part.
In midepptember 2025, Musk tweeted again
saying Grock 5 starts training in a few
weeks. Let's do the math here.
If training began in late September and
Musk is promising a release before the
end of 2025, we're looking at a
development cycle of about 3 to 4 months
from training to deployment.
Think about what that means. Gro 4
launched in July 2025.
Gro 5 is expected to debut by the end of
2025.
That's potentially a 5-month gap between
major releases, faster than any previous
cycle. And we're not talking about minor
updates here.
Based on what we're hearing, Grok 5 is
shaping up to be a generational leap
that could redefine what we expect from
AI models.
So, what exactly is XAI building? Let's
dive into the rumors and leaks because
some of this stuff sounds straight out
of science fiction.
Rumored features. Multimodal madness.
The first major upgrade everyone's
talking about is multimodal reasoning
and generation.
Now, Gro 4 already has some impressive
multimodal features. Its Grock imagine
tool can generate images and even short
videos, but industry reports are
suggesting that Gro 5 will take this to
an entirely different level. Here's what
the leaks are saying. Gro 5 is expected
to address major inefficiencies in
visual and video data processing that
earlier models struggled with. What does
that mean in practice? We're talking
about better image understanding that
could finally close the gap with GPT4V,
an integrated video generation that's
not just an add-on feature, but built
into the core model. And this isn't just
speculation. Musk's hiring patterns and
XAI's investments support this. They've
been recruiting chip experts from
Tesla's dojo team, people who understand
how to build custom hardware for AI
inference.
That tells us they're not just improving
the software, they're optimizing the
entire stack from silicon to application
layer.
Leaked benchmarks suggest Gro 5 will
have a major focus on multimodal
reasoning.
But here's where it gets really
interesting. This isn't just about
understanding images or generating
videos. We're talking about true
multimodal reasoning where the model can
seamlessly combine visual, audio, and
textual information to solve complex
problems.
Imagine describing a complex engineering
diagram verbally, having the model
understand it visually, generate code
based on what it sees, and then explain
the solution back to you in natural
language.
That's the kind of capability we're
looking at.
scale and compute the Colossus
advantage. Let's talk about the
infrastructure powering all of this
because this is where XAI's advantage
becomes clear.
Gro 5 is expected to leverage XAI's next
generation Colossus 2 supercomput. And
when I say massive, I mean this thing is
rumored to break the gigawatt power
barrier. To put that in perspective, Gro
3 already trained on roughly 200,000
GPUs. Gro 5 may use significantly more
data and compute than that. The outcome,
broader knowledge, better real-time
capabilities, and the ability to process
live data from the Xplatform instantly.
One report specifically notes that Gro
5's development targets AGI by using
200,000 plus H100 GPUs with a road map
of real-time data integration. Now
whether it actually achieves AGI is
another question, one we'll get to in a
minute.
But the scale of this training run is
unprecedented even by today's standards.
And here's what that scale unlocks.
More compute doesn't just mean a smarter
model. It means a model that can handle
exponentially more complex reasoning
chains, maintain coherence over longer
interactions, and potentially exhibit
emergent capabilities that smaller
models simply can't reach. We've seen
this pattern before with GPT3 to GPT4,
but XAI is pushing the boundaries even
further. The AGI claim, hype or reality.
Okay, let's address the elephant in the
room. In October 2025, Elon Musk stated
that his estimate of the probability of
Gro 5 achieving AGI is now at 10% and
rising.
He even went as far as to say Grock 5
will be AGI or something
indistinguishable from AGI. Now, I know
what you're thinking. Elon's known for
optimistic projections. And AGI,
artificial general intelligence, is a
term that gets thrown around a lot in
the AI community, often without clear
definitions.
But here's what XAI is actually building
toward and why these claims might not be
as outlandish as they sound.
Gro 5 is expected to incorporate new
reinforcement learning and multi-agent
training techniques specifically
designed to handle extremely complex
multi-step tasks. We're not just talking
about answering questions anymore. We're
talking about a model that can work
through hard problems more like a human
PhD researcher than a basic chatbot.
Some commentary on Grock 4.2, an interim
model that XAI has been testing,
suggests that the architecture is
focusing on longer chains of reasoning,
iterative problem solving, and agentic
use of tools. Gro 5 is expected to
continue and amplify this trend. What
does agentic use of tools actually mean?
It means the model doesn't just have
access to tools. It can decide when and
how to use them strategically.
It can break down a complex task, use a
code interpreter to solve one part,
search the web for missing information,
synthesize everything together, and then
verify its own work. That's getting
closer to how humans actually solve
problems.
Will Grock 5 be true AGI?
Probably not in the sense of general
intelligence that matches or exceeds
humans across all domains.
But could it be indistinguishable from
AGI for most practical purposes?
That's actually plausible. And if XAI
pulls this off, it could fundamentally
change how we interact with AI systems.
Context, memory, and personalization
to, but I think it's going to be
absolutely crucial. Each Gro iteration
has dramatically increased its context
window. Gro 4's API can handle 256,000
tokens and they even launched a Gro 4
fast option in September 2025 with a 2
million token context window. Let that
sink in for a second. 2 million tokens.
That's roughly equivalent to about 1.5
million words or approximately 3,000
pages of text. You could feed it
multiple textbooks, research papers, and
code bases simultaneously and it would
maintain coherence across all of it.
But context length is only part of the
story. Gro 5 is rumored to push this
even further while also refining
persistent memory features. XAI has been
developing these projects capabilities
that allow the model to remember past
interactions and maintain context across
multiple sessions. The rumors mention
sharper long-term memory and custom
personalities.
What does that mean in practice? Imagine
an AI assistant that actually remembers
your preferences, understands your work
style, recalls previous projects you've
discussed, and adapts its communication
style to match yours, not just for one
conversation, but across weeks or months
of interactions. This is what transforms
a chatbot into a genuine personal
assistant. And combined with Gro 5's
other capabilities, we're looking at a
system that could fundamentally change
how knowledge workers operate. native
tools and multi-agent systems. Now, we
get to one of the most technically
impressive aspects of Gro 5's
development, native tool use and
multi-agent systems.
Gro 4 was already trained to use tools
like a code interpreter and web search
natively, not as external plugins, but
as core capabilities built into the
model's training. Grock 5 is expected to
take this concept and run with it.
Analysts are speculating that it could
call multiple APIs simultaneously, run
code internally, and effectively
function as a multi- aent AI system.
One blog specifically posits that Gro 5
will enhance the multi- aent system
introduced in Gro 4, allowing the model
to coordinate several specialized sub
aents for different tasks.
Here's why this matters. Imagine you
give gro 5 a complex software
engineering task. Something that
requires understanding requirements,
designing architecture, writing code,
testing it, debugging errors, and
documenting the solution. Instead of
trying to do everything in one
monolithic pass, Gro 5 could coordinate
multiple specialized agents.
One agent focuses on code generation,
another on testing, another on
documentation, and a coordinator agent
ensures everything works together. This
multi- aent approach could let Gro 5
tackle highly complex software projects
or research tasks in real time, far
beyond what current chatbots can handle.
And it's not just about software.
Imagine applying this to complex
business analysis, scientific research,
or creative projects that require
multiple specialized skills.
how Grock 5 compares to previous
versions.
Let's take a step back and look at the
bigger picture. Each Grock release has
introduced a qualitatively new skill
that separated it from its predecessors.
Grock 2 brought image and video
capabilities. Grock 3 introduced think
modes for complex reasoning.
Gro 4 added native tool use and
real-time knowledge integration.
Gro 5's rumored enhancements, especially
video and audio understanding, combined
with vastly improved reasoning, would
represent another major leap forward.
And XAI's own road map hints at this
continued expansion.
The Gro 4 announcement promised ongoing
improvements in vision, audio, and
dynamic environments.
But here's what's really significant.
The cadence itself tells us something
important. Gro 5 is arriving faster than
any earlier model, roughly 5 months
after Gro 4. This signals XAI's urgency
to not just keep up with competitors,
but to leapfrog them entirely.
And when you look at the competitive
landscape, this makes sense. Open AAI
has GPT5 in development. Google is
pushing Gemini. Anthropic has claude.
The AI race isn't slowing down, it's
accelerating. and XAI seems determined
to not just participate but to lead
release outlook and what to expect. So
when can you actually get your hands on
Gro 5?
Based on all the clues we have, it's in
latestage development right now. Musk's
timeline of training soon in September
2025 suggests a launch before the end of
2025, which at this point means we're
potentially just weeks away. It will
likely debut to XAI's premium users and
partners first. That's been the pattern
with previous releases. We'll probably
first see announcements or demos on X or
XAI's blog just like we did with Gro 3
and Gro 4. But here's what I'm watching
for the first benchmarks.
How does Gro 5 actually perform compared
to GPT5, Claude, and Gemini? The real
world usability. Does the multi-agent
system actually work in practice or is
it overhyped?
and most importantly the pricing and
access. Will this be available to the
general public or will it be locked
behind premium tiers? What we know for
certain is that analysts and Musk's own
hints paint a picture of an AI model
genuinely aiming for AGI like
performance, multimodal, massively
scaled, and optimized for reasoning. If
even half of this promise comes true,
Gro 5 could be a major milestone in the
AI arms race.
So here's the bottom line. Gro 5
represents XAI's most ambitious project
yet, potentially arriving just 5 months
after Gro 4 with capabilities that could
redefine what we expect from AI systems.
We're talking about a model that
combines advanced multimodal
understanding, AGI level reasoning,
massive scale, persistent memory, and
multi- aent coordination all in a single
release.
Whether Musk's claim of 10% probability
of AGI is accurate or optimistic, one
thing is clear. XAI is moving faster and
pushing harder than almost anyone
expected.
And that's forcing every other AI
company to accelerate their own
timelines. The AI landscape at the end
of 2025 could look radically different
than it does today. And Gro 5 might be
the catalyst that makes that happen. If
you want to stay updated on Gro 5's
release and my hands-on testing when it
drops, make sure to subscribe and hit
that notification bell. And let me know
in the comments. Are you excited about
Gro 5 or do you think the AGI claims are
overhyped? I'd love to hear your
thoughts. Thanks for watching and I'll
see you in the next one.