We've got some wild AI stories heading
into this week with major model releases
and leaks starting to surface. First up,
Enthropic appears to be preparing the
return of Cloud Sonic with their fifth
variation which recently showed up as a
model slug through a partner provider.
Early testing suggests that it is a
solid upgrade with impressive outputs
and I'll be showing some later in this
video. But that's not all. Enthropics
also has seemed to have trained a model
that is even more capable than Mythos.
Yep, you heard that right, more capable
than Mythos already. These reports
indicate that a new version of Mythos
has emerged from training which is
pushing performance even further than
the already controversial model and has
been banned already. Meanwhile, Open AAI
looks ready to launch GBT 516 this week,
potentially as early as Tuesday or
Thursday. On top of that, we're
expecting a brand new voice model, and
it is a huge upgrade, the first major
refresh since the launch of GPT4 Omni.
We've also have a new player entering
the Frontier AI race. And this is a new
Japanese AI lab called Sakana. And it is
a new AI lab that unveiled a new model
that's called the Fugu and Ultra
Systems, which claims performance
comparable to Claude Fable 5 and Mythos
5 across number of benchmarks.
Unfortunately, after taking a closer
look at the results myself, those claims
appear to be a bit optimistic. Sorry to
burst your bubble, but it's nowhere near
that level yet. But it's still pretty
good. Now, that's just the beginning.
So, let's actually go ahead and take a
look at everything that is happening in
the AI space right now. If you want the
best AI tools, workflows, and drops
before everyone else, join my free
newsletter with the link in the
description below, which is completely
free. Starting things off with Claude
Sonnet 5 because the Cloud Sonnet 5
model slug has appeared in Enthropics
partner provider which historically
means that if it starts showing up as a
model slug within these different
partner programs or providers they
usually appear around 5 to 7 days before
it actually ends up launching which
would hint at it dropping this week. So,
if this pattern holds, we will
definitely be having a very busy week
because we're going to see the Sonnet 5,
which will be dropping very soon. This
is a pretty big deal because Sonnet is
Enthropic's main daily driver model. It
is the model that most people actually
use for coding, writing, agents, and
general workflows. And it has been quite
some time since it has gotten any sort
of update. With Sonnet 5, we should
expect a much larger and improved
context window, potentially around one
to two million tokens, better vision,
stronger understanding of UI mockups,
and better handling of architecture
diagrams. Now, there are also signs that
Enthropic may be using a new tokenizer,
which could use around 30% more tokens
on the same prompts. So, while that
might make prom slightly more expensive
or tokenheavy, the trade-off could be
much stronger reasoning, better
multimodal understanding, and a major
upgrade over the current Sonnet
experience. Here is the first look of
the Sonnet 5, and this is an SG of a
Nintendo Switch 2, and the model is
really fast in terms of outputting a
qualitative design of any sort of SVG,
which you see over here. And this is
where it wasn't provided any reference
image. It was able to solely create all
of this all on its own. And that is
quite impressive because this is an SVG
generation of a device which truly looks
like it's an actual image. On the topic
of Enthropic, this is where things get
really interesting cuz a new and more
capable version of Mythos has reportedly
emerged from training. Now, we don't
know if Enthropic is going to call this
Mythos 5.1 or Mythos 6 or if they're
even going to end up releasing this, but
the important part is is that it has
arrived. Now, this is a reminder that
even when a model is not being served to
the public, that does not mean
development has stopped. Stopping models
like Fable 5 or Mythos 5 from being
publicly available does not slow down
labs internally. In fact, it may even
speed things up slightly because instead
of using compute and resources to serve
public users, those resources can be
redirected towards training, testing,
evaluations, and the next generation of
models. So, I know Mythos 5 Fable 5 has
been banned right now, but there are
internal steps that these companies are
taking to measure what sort of guard
rules should be in place to avoid any
sort of future bans. Now, regardless of
that, they did already ban this new
variant, which is could be the Mythos 6
or whatever the naming may be, but
there's no rules preventing that these
frontier labs can continue pushing
capabilities even further with models.
It's just publicly facing models can't
be accessible to everyone just yet if
they are this capable and powerful. But
according to Curran which is a reputable
source on enthropic leagues essentially
he had stated that it is going to be a
model that is truly stronger much more
stronger than what we had saw with
mythos and that is just incredible if it
is taking a major leap from mythos but
it is likely better at longer horizon
reasoning stronger agentic coding
improved planning and more reliable
execution on massive tasks. Biggest
question though is that if Enthropic
will release it through project glass or
turn it into future public models or
keep it behind the scenes to accelerate
their next generation of systems. As we
all know, Open AI is also expected to
launch GBT 5.6 Pro this week, probably
on Thursday. And alongside that, there
seems to be a new family of models that
they're also going to be launching. Now,
we've already tested GBT 5.6 6 Pro quite
a lot and we even made a video which you
can actually take a look at in the links
in the description below. And the model
definitely feels quite special with the
right prompts and it can basically build
anything. It has also gotten
substantially better at front end. Its
design taste has gotten a lot better and
on top of that open AI is also preparing
a new voice model called BD. This is
GBT1
and it is something that has an August
2025 knowledge cutoff which is the kind
of voice upgrade people have been
waiting for since GPT4
Omni or for Omni GPT4 Omni. The earlier
GPT 5.6 versions that we were testing
were under the checkpoint called Kindle
Alpha, but we should expect newer
checkpoints before a final release. But
on the topic of this new voice model,
this is a new voice mode that will be
implemented into chat GPT powered by the
BD model. And this has actually already
started rolling out to some users within
the chat GBT app. And it is definitely a
major step forward from traditional
voice assistants because the model can
interact with you in real time instead
of simply waiting for you to finish
speaking. Think of it like Gemini's live
mode, but not at that level yet. And
it's not as multimodal as Gemini. It can
interpret naturally. It can count as you
speak, follow along during a
conversations, and even correct you when
you make a mistake mid-sentence.
Overall, it is definitely a better
conversational model, and it feels much
more fluid and humanlike, which is going
to make Chat GBT a better real-time AI
assistant. So, this is definitely
another nice quality of life update.
Just take a listen to what it actually
sounds like.
My knowledge cut off is August 2025. If
you need something more current, I can
check for updates.
>> Like a paragraph and then just tell me
like the name of the food I say when I
say it and the counter like count how
many foods I say.
>> So
>> yeah. Yeah,
>> let's go. Uh so yesterday I was just
sitting on the couch minding my own
business when my phone buzzed like 17
times in a row and I swear my heart be
banana flip because I thought something
really was actually wrong. Turns out it
was just my
over some random meme and I felt so
silly that I nearly threw my mango for
pain out the window.
>> Anyways, I
for 10 minutes and somewhere in there I
remember that I still had like a whole
apple of homework sitting untouched on
my quietly.
>> Got it. You mentioned three foods in
total. Banana, mango, and apple. Each
said what?
>> Going back to the GBT 5.6 Pro leaks.
This is where things are getting even
crazier cuz one of the latest tests show
a model generating a complete firsterson
playable interior house which is being
done in a single HTML file. It took
approximately 40 minutes and the GBT 516
Pro was able to create this in a 700 KB
HTML project with multiple rooms. You
have a coherent floor layout. You have a
detailed bathroom, interior navigation,
and a smooth firstperson movement system
built within this generation, which is
definitely remarkable. And this is why
people are starting to say this model
feels different. It's not just
generating basic web pages or small
demos anymore. It's much more
intelligent. It is able to play through
intelligently based off of the prompt
and use creativity and be less lazy in
terms of outputting the contents that
you're looking for. And based off the
leaked testings, we're definitely going
to see this release being talked about
on the same level as Mythos 5. Lastly,
we have Sakana AI, which is a new
Japanese AI lab that announced Sakana
Fugu and Sakana Fugu Ultra. And these
are not normal natural language models.
The fugu is more of an orchestration
system that is trained to operate other
large singish models similar to how open
router has the fusion API. But in this
case, it means that it can route tasks.
It can combine different models and use
multiple systems together to get better
results. And according to Sakana, the
Fugu and Fugu Ultra perform on par with
models like Claude Fable 5 and Mythos 5
across several benchmarks. And I do
think these claims need to be taken with
caution because benchmark performance
does not always translate into real
world capability, but the trigger point
is that it's probably where AI is
actually heading. And based off my own
tests, I believe this model is nowhere
near Fable 5 or Mythos 5. If you like
this video and would love to support the
channel, you can consider donating to my
channel through the super thanks option
below. Or you can consider joining our
private Discord where you can access
multiple subscriptions to different AI
tools for free on a monthly basis plus
daily AI news and exclusive content plus
a lot more. Now Sakana Fugu Ultra
against the Claw Opus 4.8 Ultra Code is
pretty interesting cuz on 3GS Crossy
road style game generation test. This is
where Opus delivered a better final app
in terms of design, functionality, and
polish. But it took 79 minutes, around
940K tokens, and costed around $37.85
and got stuck twice. But Fugu Ultra had
issues like inverted controls, a wonky
camera, and no FS now SFX, but it
finished it around 22 minutes. So, it's
much faster and only around 90k tokens.
But the cost was drastically cheaper,
$7.32,
and had way better difficulty scaling.
So, Opus won quality, but Fugu won
speed, cost, and efficiency. But that's
about it, guys, for today's video on all
the latest AI news. Make sure you go
ahead and subscribe to the second
channel, join our newsletter, join our
Discord, follow me on Twitter, and
lastly, make sure you guys subscribe,
turn on notification bell, like this
video, and please take a look at our
previous videos so that you can stay
upto date with the latest AI news. But
with that thought, guys, have an amazing
day. Spread positivity and I'll see you
guys fairly shortly. Peace out, fells.