Kind: captions
Language: en
Gro 4 just dropped and the numbers are
absolutely insane. We're talking about
an AI that scored 50% on a test where
humans barely hit 5% trained on 100,000
GPUs and it's rewriting everything we
thought possible about artificial
intelligence. Welcome back to
bitbiased.ai.
In five minutes, I'll break down every
announcement from XAI's Gro 4 launch
event. We're covering unprecedented
benchmarks, multi-agent collaboration,
voice capabilities, and a roadmap
including AI generated movies by next
year. Let's dive into why Gro changes
everything. Revolutionary intelligence
and performance academic breakthrough.
Gro 4 achieves perfect scores on
graduate level exams like the SAT and
GRE without prior exposure. They created
humanity's last exam, a brutal 2500
problem test spanning mathematics,
physics, chemistry, linguistics, and
engineering designed by actual experts.
While humans struggled to score 5%,
Grock 4 hit 38.6%.
Grock 4 heavy, the multi-agent version,
scored 50.7%. That's 10 times better
than human performance on humanity's
most challenging academic benchmark,
Training Revolution. The timeline is
mind-blowing. Gro 2 was just a concept
12 months ago. They trained Gro 4 on
Colossus, their supercomputer with over
100,000 GPUs, 10 times more compute than
Grock 3. Unlike traditional models that
learn text patterns, Gro 4 uses
reinforcement learning from first
principles. It's rewarded for answers
grounded in logic and observable
outcomes. As Musk said, physics is the
law. Everything else is a
recommendation, native tools and
benchmarks. Gro 4 was trained to use
tools natively, not as external add-ons.
When you include tools, accuracy jumps
from 26.9% to 41%. The benchmark numbers
are staggering. 91.7% on AM25,
96.7% on HMMT25
and 15.8% on the private V2 benchmark.
More than double clawed opus. It's the
only model to break 10% on V2.
Revolutionary features available now.
Multi-agent system Grock 4 heavy spawns
multiple internal agents that solve
problems independently, share insights,
then collaboratively produce superior
answers. Think PhD level experts
collaborating at machine speed. Each
agent approaches problems differently,
converging on solutions no single agent
could achieve. Voice mode 2.0 and real
world apps. Voice capabilities are
impressive. Latency cut in half with
five new voices, including Eve with a
British accent that whispers poetry and
sings opera about Diet Coke. In blind
testing, Grock voice is ranked as more
natural than competitors. The Allen
Institute uses Gro 4 for crisper
research hypothesis. It's the top rated
model for chest X-ray diagnostics. In
vending bench simulations, Gro 4
achieved 4,694
mean net worth versus Claude's 277 and
humans 844. It sold 4569
units compared to humans 344. Game
development and API. A developer created
a complete firstperson shooter in 4
hours using Gro 4, which autosource
textures, models, and game logic. The
API is live with 256 kilobyte context
windows, function calling, and real-time
X research capabilities.
Current limitations. Gro 4 is currently
weak in image understanding, described
as squinting through glass. However,
foundation model 57 is training now due
in August, which will boost image
comprehension, video analysis, and
audiovisisual reasoning. The future
roadmap and pricing August to October
timeline. August brings a dedicated
coding model for production quality code
generation. September delivers
foundation model VI7 with full
multimodal agents processing language,
images, audio, and actions together.
agents that can understand your screen
and execute complex workflows
independently. October launches AI video
generation using 100,000 plus GPUs. The
timeline quarter 3 2025 for AI generated
video, quarter 4 for AI television and
2026 for fully AI generated films that
script, animate, and render end to end.
Access and pricing. Super Gro costs $30
monthly. Super Grock Heavy is $300
monthly. Both provide access to Gro 4
and Gro 4 heavy with higher rate limits
and early access to new features.
Conclusion:
Gro 4 represents a fundamental shift in
artificial intelligence. From PhD level
reasoning across all subjects to multi-
aent collaboration, from native tool
integration to AI generated movies, XAI
isn't just improving AI incrementally.
They're redefining what's possible. The
most striking aspect isn't any single
capability. It's the speed. 12 months
from concept to world's smartest AI.
That pace suggests we're entering
exponential advancement that makes
recent progress look slow by comparison.
What impressed you most about Gro 4?
Drop your thoughts in the comments and
subscribe to bitbiased.ai
for unbiased coverage of every major AI
development as it happens. Thanks for
watching.