I'm going to be vibe coding with the newly released Claude

sonnet 4.6 This was a model that released yesterday from

anthropic and so far.

I am very impressed I was able to use this model

extensively yesterday and so far.

I have a lot to say about it I've already put it through

the bridge bench I've put it through the creative HTML test

and I have a lot to talk about and show you in this video

We are going to be looking at benchmarks We are going to be

putting it through production vibe coding tasks to really

give this model a fair shake And by the end of this video,

I will share whether or not it is bridge mind certified

with that being said if you haven't already Liked

subscribed or joined the discord make sure you do so

There's a link in the description down below to join the

fastest growing vibe coding community on the internet right

now And with that being said,

let's get right into the video All right So the first thing

that I want to do is actually take a look at this model

inside of open router see how fast it's running From

anthropic check out the price and compare it a little bit

with opus 4.6 So right away you can see the million context

window.

That's pretty that's pretty good,

but that's was expected from this model We also can see a

three dollars per million on the input in a fifteen dollars

per million on the output That is the same exact pricing as

sonnet 4.5 If we compare this now to opus 4.6,

you know You're talking about a 1 million in context with

opus 4.6 and then the price Five dollars and twenty five

dollars respectively.

So you're talking about like close to a 40 What is that 40%

cost reduction when talking about sonnet compared to?

Opus so you're talking about a 40% cost reduction The

question really is is okay How much of a speed up is sonnet

compared with opus and we can see that here So here you can

see that from anthropic anthropic is putting out this model

at 40 tokens per second And with opus you're looking at 32

tokens per second So what even is that a 25% speed up I

believe so you're talking about a 25% speed up when when

comparing sonnet?

Versus opus which is pretty good get 25% faster And then if

you look at the latency the latency 1.22 seconds versus 2

.25 seconds,

so it almost cuts the latency is half So this is a more

performant model for sure and then you're talking about a

little bit more of a price reduction But something I do

want to emphasize is that if you guys remember Claude

sonnet 5 4.5 was very very popular and The reason that it

was popular is that at the time that Claude sonnet?

4.5 was the best model that anthropic had it was when opus 4

.1 was the leading opus model Okay And something I want to

highlight here that some people may forget is that the

reason that sonnet?

4.5 was so much better at that time,

which is what I don't know maybe four or five months ago at

that point The reason it was so much better is that it had

this price three dollars per million and fifteen dollars

per million While the leading opus model was fifteen

dollars per million and seventy five dollars per million So

this was a 5x in an increase in price to use opus models

But that's no longer the case right and the point that I

want to make here that everybody needs to understand is if

you guys Remember when sonnet 4.5 was like by far the best

model that everybody was using that was when opus was way

more expensive And they really reduced the price for opus

So that makes it a little bit more that brings in the

debate it makes the debate a little bit hotter for hey

Should you use sonnet or should you use opus?

So with that being said,

let's now take a look at the benchmarks and look at the

improvement from sonnet 4.5 to 4.6 So here it is.

You can see the terminal bench 2.0.

You're looking at an 8.1 percent improvement agentic coding

suite bench verified This is the most in courting important

coding benchmark that I always look at and here they were

able to get a 2.4 percent Improvement which is okay,

but you're talking about a one point.

It's still 1.2 percent worse than opus 4.6 So in terms of

coding you're still looking at that opus 4.6 and saying

okay This model is more capable there now if you do compare

this in the agentic computer use It's just right up there

with opus 4.6 It does very good with computer use and if we

actually go to the blog that anthropic put out They

emphasized computer use a lot in this blog.

Look at this.

They actually emphasized this first They said almost every

organization has software It can't easily automate

specialized systems and tools built before modern

interfaces like API's existed to have AI view software

Software users would previously have had to build bespoke

connectors But a model that can use a computer the way a

person does changes that equation So this is something that

I've talked about previously on on live streams and in my

channel is just the fact that Hey computer use is coming

That's the reason that I think open claw is such a big deal

is that computer use is going to become a much bigger deal

In the future you're going to have isolated environments

and these AI agents are going to be using these computers

Look at sonnet 3.5 in October 20 24 Compared to where

sonnet 4.6 is now look at that 14.9 compared to 72.5,

which is just insane So take note of that also They did put

out look at this the agentic financial analysis that beat

out the office tasks It also beat out it did perform very

well an agentic tool use right up there with opus 4.6 So

this model is on par with opus 4.6.

It's not it's not better than it You know,

we've seen sonnet models like sonnet 4.5 was better than

the opus model that existed at the time You're like, okay,

use use sonnet 4.5, right?

But now that we have opus 4.6 with the price that we're

getting it at,

you know You're only talking about a 40% price reduction

and then a 25% speed improvement So that does kind of bring

into the equation the equation What should you be using

look at novel problem solving opus 4.6?

10% higher 10.5 percent higher than sonnet 4.6 So there's a

lot that you can look at here and say okay opus 4.6 is

still a really capable model That is better than sonnet 4

.6.

So you have to weigh Okay,

do I want that improvement in speed but do I also want to

have that at a price reduction, right?

So like you have to you just have to ask yourself that

question Another thing that they emphasized is that users

preferred sonnet 4.6 to opus 4.5 Their frontier model from

November 59% of the time.

This is really interesting that they said this So 59% of

the time people actually preferred this model and they

relate they rated it as significantly less prone to Over

-engineering and laziness.

So that's very very interesting.

It means that it's better instruction following Okay,

so that's another important thing.

Also look at this guys So if you guys are familiar with the

vending bench,

this is a really trending bending Benchmark the vending

bench.

Look at how it performed sonnet 4.6 outperforms sonnet 4.5

on vending bench arena by investing in capacity Early then

pivoting to profitability in the final stretch.

Look at how it performed 6,000

versus like barely over 2,000

So it was it was like close to 3x more performant,

which is insane So you're looking at this and you're saying

okay Look at the difference between sonnet 4.5 and 4.6 Like

that's actually a pretty noticeable difference And what we

want to look at is we now want to look at the bridge bench

and we want to see okay How does this model actually

perform in our own personal benchmarks?

And what we've done is we've created bridge bench to do

that and I want to emphasize that I have Now index this in

bridge bench.

You guys can go check it out If you go to bridge mind and

just go to community go to bridge bench There's also going

to be a link in the description down below but check out

how this performed.

So Here's Claude Opus 4.6.

Here's Claude sonnet 4.6.

Here is the entire ranking so you can see here's GLM 5

Here's Quinn 3.5.

Here's mini max m 2.5.

Here's GPT 5.2.

Codex So Claude sonnet 4.6 outperforms GPT 5.2 codex.

It does better in completions.

It also is faster So here you can see that the the total

time and I'm gonna change up the formatting of this But the

total time for sonnet 4.6 to complete this benchmark was

924 seconds Whereas with Opus 4.6,

it was 983 seconds and with GPT 5.2 codex.

It took 2245 seconds,

so that's where you know with with these philanthropic

models There's so much more performance and speed and I

will say that they did improve this with GPT 5.3 codex They

improved the speed but this model we can't benchmark it yet

since it's not available in OpenRouter or the API The next

thing I want to take a look at is the creative HTML

benchmark So this is a benchmark that we use to be able to

actually look at UI comparisons So the first one is the

lava lamp So let's go take a look at the lava lamp that

sonnet 4.6 produced.

Here it is right here So this is the lava lamp produced by

sonnet 4.6.

You guys can make your judgments on this Let's go back over

to Opus 4.6 and look at this.

So here's Opus 4.6.

Here is sonnet 4.6 So definitely different approaches.

I don't know what you guys think is better But here's the

lava lamp.

Another one that I like to take a look at is the retro

space invaders This is probably my favorite one to look at.

Here's how Opus 4.6 performed I think it did very well on

this on this creative HTML test,

but let's take a look at how sonnet 4.6 performed Let's

take a look at it here sonnet 4.6 And I did post this on X

and a lot of people were saying that they thought that

Sonnet 4.6 did a better job because it followed the

instruction look at the alien behind it like added in a

boss It added in a really just the spaceship down there at

the bottom is a little bit better styled It did make it

more retro.

So you're seeing the styling and the styling so far is

pretty impressive So in terms of styling,

I think that this is a very capable model for styling Let's

go check out if it is in design arena.

Let's go over here and check out design arena AI let's go

see if it's in design arena yet.

I actually don't know if it's in design arena Let's check

the leaderboards though.

So sonnet it is in design arena.

Look at this.

So sonnet 4.6 performs 1338 so it is behind opus 4.6.

So opus 4.6 got 1397 the thinking version got 1373 and then

GLM 5 got 1354 and then sonnet 4.6 Performed for they got

fourth place with 1338.

So that's actually interesting It did perform worse than

GLM 5 but still up there highly outperforming Gemini 3 Pro

and Way outperforming GBT 5.2 high so that gives you guys a

look at the UI and the styling There I think that even

though like it is 1338, you know,

you are bringing into the account of like,

okay It is more affordable.

It's faster.

That is very important to note Also another important

benchmark that I do want to pull up is the artificial

analysis This is one of the most important benchmarks that

I look at.

So let's see.

It should be benchmarked Let's take a look and just make

sure real quick.

So here is the artificial intelligence.

This is a really important benchmark Let's see if it's

indexed here.

Let's go to sonnet 4.6 it is here.

So here it is adaptive reasoning and then let's also add

GBT 5.2 to the mix as well I think I turned it off by

accident.

So let's see here sonnet.

Oh my gosh guys Look at this sonnet 4.6 max is ranked

number one.

That is insane.

So it actually outperforms G Or I guess it's Thai,

but I guess it's outperforming here.

Maybe it did a little bit better So it's outperforming but

look at this sonnet 4.6 on the intelligence index

outperforms opus 4.5 And let's take a look at opus 4.6 and

add that to the equation.

So here's opus 4.6 So it is under opus 4.6 And again,

that's what we're seeing is we're seeing whereas with

sonnet 4.5 versus opus 5 or opus 4 Sonnet 4.5 was a better

model at that time, right?

But what we're seeing is that opus 4.6 is actually still

Outperforming sonnet 4.6,

which is actually a good thing if we go to the coding

index.

Let's take a look there Wow, look at this guys.

So on the coding index,

this is big So on the coding index sonnet 4.6 is the best

performing model in the artificial intelligence analysis

coding index Look at this 49 and then 51.

So this is the highest performing model in the artificial

analysis coding index I definitely need to make an ex post

about that because that's a big deal.

This this is a high value Index so we want to take a look

at this now.

Oh my gosh Agenic index it absolutely performs really well

Let's add is GLM 5 added to this index as well because I

know GLM 5 performs very well on this one as well Let's add

it so GLM 5 performs well on this agentic index But look at

this it destroys GPT 5.2 and again a little bit under opus 4

.6 max But look at that 63 that's very very performance.

Another thing that we want to take a look at is the

hallucination index So let's scroll down a little bit and

take a look at that.

That is the index that measures the hallucination rate So

let's take a look at this.

Wow.

Wow chap.

Look at this 38% on the hallucination index This is insane

So if you look at opus 4.6 That's at 60% and on this

benchmark the lower the score the better and look at that 38

% so that is a 10% reduction in the hallucination rate from

sonnet 4.5 So that is very very impressive.

That's something to look at right hallucination rate is

very very important You know a big problem with Gemini

models is that they hallucinate all over the place, right?

Like or even is Gemini 3 Pro look at that 88% in something

that we see with these Anthropic models is that they do

have very good hallucination rates, right?

Meaning lower hallucination.

You look at GPT 5.2 extra high 78% where now look at this

sonnet 4.6 38% So this model is going to hallucinate less

than opus 4.6 and that is going to make it more reliable So

you're talking about a model that's going to be more

reliable.

That's faster.

That's cheaper This is this is this model has a lot of

potential guys and that's going to summarize the benchmark

review And now I want to move on to actually using this

model inside of quad code and talking to you guys about

what I saw Yesterday in my session my vibe coding session

at a vibe coding No talking stream and I also was using

this model when it initially launched during my vibe coding

in app until I make a million dollar Series so with that

being said,

let's launch Claude code and actually start vibe coding

with this model All right guys,

so I have bridge voice and bridge space launched These are

two tools in our vibe coding suite that I'm going to be now

using in order to vibe code with sonnet 4.6 You can see

bridge voice is a very good voice to text tool and a bridge

space is a great ADE agentic development Environment to be

able to manage an agent swarm So you can see I have 16

sonnet 4.6 agents open and I have my voice to text tool And

I'm just gonna start working like I normally would in

production and you guys can follow along as I put this to

the test But for a minute here,

I'm just gonna put all of this to the test to watch this

The first thing I'm going to do is drop in bridge bench I

need you to review a bridge bench and take note of the new

speed test bench that we have What current results do we

have coming out of the speed test?

Are there any models that have completed the speed test?

So we're going to drop this in and then we're also going to

go over here and we're going to drop in the bridge Mind UI

I need you to do an in-depth review launch three sub agents

to be able to review this website and review it in terms Of

the performance for lighthouse and make sure that this

website is optimized for speed and performance as well as

for SEO Do an in-depth deep dive and analyze it and then

tell me your findings with these three sub agents So I'm

going to drop these in here.

I'm then going to go over here.

I'm going to drop in a bridge voice.

I Need you to do an in-depth review of the customized

scroll bar that is in this application I need you to make

sure that it has an on theme scroll bar for this app that

works Well both on Windows and Mac and Linux I also need

you to make sure that that top header looks good and is

customized for Windows Mac and Linux So let's now do that.

So it looks like only one model has the results so far

anthropic So what I need to do now is I'm going to drop in

bridge mind UI I need you to review this website and go to

the bridge bench page and add this as a toggle for the

speed test bench Where we will be able to display the

leaderboards for the fastest models Please create this new

toggle in the bridge bench page and then update accordingly

add this sonnet 4.6 model to the leaderboard there So we're

going to pass this in and just for reference So you guys

can see what that is doing So if we go over here and we

actually go over to local host What you guys are gonna see

is that we can go to bridge bench,

which I kind of show you guys earlier but you can now it

art Wow it already added the speed bench and Wow,

so it already did that that's actually really impressive So

let's um,

let's now go over here and let's let's drop in another

prompt.

So let's drop in bridge voice I do have an issue with

bridge voice locally I need you to launch three sub agents

to do an in-depth review of bridge voice We made some

updates recently when we added in the ability to be able to

toggle the view of the widget between Logo only versus

having the logo and bridge voice for the widget pill and

now the widget pill is no longer showing when I initially

launched bridge voice would you launch three sub agents to

do an in-depth deep dive to figure out why the Widget pill

is no longer displaying on Mac Do an in-depth deep dive and

figure out what is causing this and then implement a fix so

that it does display on launch So I'm gonna do that and I'm

gonna launch another agent and this one is for bridge

Actually,

I can't do that because we're using it locally right now So

let's see this one.

So these agents are all working right now This one is doing

a look there and then I'm also gonna do a follow prompt on

this one I need you to look at the models that we have

tested in open router like GLM 5 Minimax M 2.5 and I now

need you to launch sub agents for each of these models that

we have not put through the speed Bench and I need you to

benchmark mark these models and then output their results

respectively But launch sub agents for each model that we

have not yet put through the speed bench yet So we're gonna

put all those models.

I'm just gonna send that there we're also going to launch

here and I will say that so far what I am finding is that

this is Noticeably like I was using this last night and

this model is very fast and it is reliable You know when

you look at that hallucination rate one thing that I can

say for certain is that it is more it is very reliable It's

a very reliable model.

It's cheap.

It's fast.

I need you to create an agent team that is built around

identifying AI slop code and Ways that we are able to

improve performance in bridge mind API You need to create a

team that will be able to evaluate the API which is a nest

JS Application and this team will be able to work together

to identify findings that we can improve the code without

breaking functionality Improve performance without breaking

functionality launch this team and output your findings and

then using that same team I need you to implement fixes

again without breaking any functionality.

So let's also paste that in So that is going to create a

team So we'll put teams to the test and one thing with

teams is that you know quad code they released teams and

teams with something I said,

hey This is going to be useful But Claude Opus 4.6 is just

a little bit too expensive to actually get the use out of

teams So with sonnet the interesting part about sonnet is

that sonnet is a 40% cost reduction So with that 40% cost

reduction that is going to make teams a little bit more

usable because with Opus if you launch a team It's going to

max out your usage immediately.

But with sonnet, it's you know,

you're looking at 40% cost reduction So you're just gonna

get way more use out of your plan,

right and it's right up there in performance It's even

better with the hallucination rate.

So definitely something to take note of let's launch bridge

my API and I already put it in I need you to run the test

suite and fix any failing tests.

Let's launch this one.

So you can see that team is now being created Let's now go

over here and drop in another one here and I'm gonna drop

in the bridge mind web app and the bridge mind UI I need

you to do an analysis of the onboarding flow and I need you

to make sure that the styling is polished and that it Is

accurate and flows well with best practices.

So let's drop this in another interesting thing that we

have is this one is now done So this one is giving us about

the lighthouse performance 72 to 78 the target is 90 plus

So total estimated our 30 to 40 hours for full

implementation.

So it's not at 4.6.

Just probably take about 10 minutes So I now need you to

output all of the findings so that we can get our

lighthouse score to improve Run the test and then create a

structured readme file with everything that is needed So

that we can create documentation and then be able to start

improvements But first you need to create this readme you

need to run the test and then create the readme

respectively Okay,

so let's paste that in now and then I'm gonna go over here

and I have an interesting problem here Which is the URLs

when I'm redirecting with bridge space and bridge voice.

So check out the URLs here So you see this one redirected

to this This here this port and then this desktop code It's

not a very friendly URL and what I want to do is I want to

improve the URL redirects for both bridge space and Bridge

voice without impacting the auth functionality.

So I'm gonna drop in both of these redirects and I'm going

to paste this in I'm going to say I want you to take a look

at the URLs that I get sent to when Using the auth flow for

both a bridge space and bridge voice take a look at these

URLs And then also can I get this can I go back one?

So yeah I want to go back one too so that it can take a

look at like the other step Because I think that this can

be improved and this is a little bit more of a just

difficult and nuanced task It's a little bit weird, right?

So let's drop this in so we have bridge voice bridge space

and you can see that even in the bridge Voice it says

electron redirect, right?

That's not necessarily correct, right?

It's a little bit outdated So let's now pass in bridge

voice and let's also pass in bridge space tari And what

we're now going to do is we're going to say I want you to

do a deep dive on the redirect URLs and the auth flow for

both bridge space and bridge voice think through best

practices for creating a user friendly URL system you can

see that for bridge voice it has electron redirect even

though this even isn't even an electron application So

there's some issues there It would be really nice if we

could optimize the URLs to be user friendly and follow best

practices launch three sub agents and Search the internet

for the best practices for this and then I need you to

create a structured plan for updates That will not break

functionality But we'll update the URLs and this off flow

to be to have URLs that are more user friendly and follow

best practices So I also need to drop in the bridge mind UI

and again We are currently using bridge voice and bridge

space which are tools in our vibe coding tool suite You can

currently subscribe to bridge mind pro for 50% off your

first three months So $10 a month for your first three

months,

you can go to bridge mind AI and learn more But these

products are getting closer and closer to shipping They

already are in production for Mac OS and Windows and we're

getting to more stable releases But these products are

going to be very helpful for vibe coders and I use them

daily in my workflow So it's a creative match patterned

here.

So this is the recommendation So these are all the issues

that it found So I'm now I want you to launch sub agents

that will implement updates for each in the best practice

way That will not break functionality,

but we'll implement the fixes accordingly So we're going to

now pass that in it looks like all tests of two thousand

two hundred eight tests are all passing So we don't need to

worry about that.

Let's see this here.

So we are passing these tests It looks like let's let's

look at the speed test here.

So Aurora alpha two hundred fifty eight tokens per second

cloud sonnet four point six 63 tokens per second.

So here on the speed test this actually was running at 63

tokens per second So even though it said 40 tokens per

second in my speed test I got sixty three point eight

tokens per second,

which is absolutely nuts inside of open inside of open

router inside of the bridge bench Speed test.

So that's one thing to look at there.

Sixty three tokens per second I'm very interested to see

what we get out of Opus four point six It looks like Quinn

three point five was at fifty one and then a couple more

are still running I need you to add Opus four point six to

this as well.

So let's let's do this as well We'll get the results from

Opus four point six here These agents are analyzing and

what I will say about using this model is what yesterday I

was using it I'm using it right now.

And what I will say is I notice it being more reliable I

noticed the hallucination rate being lower I notice it

being faster and I notice it being able to one-shot So in

terms of talking about this model and whether you should be

using it whether it has the bridge mind stamp of approval

Boom bridge mind stamp of approval.

I will say I can tell that this is a good model just after

using it This is something that you want to start

implementing in your workflow that 40% cost reduction and

the speed increase for this model is very impressive It's

noticeable.

You're going to notice it when you're launching teams,

you're going to notice it when you're using sub agents You

know if you go to this is another thing that I wanted to

talk about if you go to model, right?

This is something that you guys need to know about.

So the model menu has changed for quad code So check this

out.

So select model switch between cloud models.

So the default recommended is Opus four point six This is

the most capable model for complex work and you now have

the ability to this is this is new with yesterday's

deployment Of sonnet four point six.

They also shipped this so now you can choose the one

million in context for Opus four point six But look at

this.

It's billed as extra usage.

Do you guys see this?

So I don't know why they necessarily did this This is

weird.

So They have the default right?

But then they also have the one million in context and then

that's billed as extra usage and I don't know why they did

that So if you want to get absolutely taxed you can select

the one million in context and then it's the same thing for

sonnet So you can choose sonnet and it says best for

everyday tasks and then you can also choose sonnet 1m and

it said billed as extra usage and This is actually billed

at a higher price.

So six dollars per million and $22.50 per million,

you know,

that's that's something interesting and it's a little bit

misleading If I'm being completely honest because you're

looking at open router, right?

And if you go over to open router What we're gonna see is

you look at it and you're like,

oh one million in context three dollars per million on the

input Fifteen dollars per million on the output.

That's great But then in quad code for some reason they're

making these 1 million in context models Basically unusable

because they're not even allowing you to use it with your

plan.

They're billing it as extra usage So this is something I

wanted to highlight Maybe you guys can let me know in the

comment section down below is the like what is this model

even running at right?

Like sonnet, can I even ask it?

Like could I switch over here and could I say what is the

context window on this model?

Will it will it respond?

Will it let us know because 1 million in context is like

great, right?

Wow Okay,

this is actually something new that we just learned guys So

sonnet 4.6 if you're using it just normally without having

it set to this 1 million in context It has a two hundred

thousand context window that is actually a little bit.

I'm gonna dock it a little bit here This is not good.

This is not a good approach.

So they're billing it for a higher price with more cost per

token you're unable to use the 1 million in context with

your plan and the normal default mode only has a 200,000

in context like that's really really interesting because if

you look at Gemini obviously Gemini models They all have 1

million they have over a million in context and this is

just hey if you're using Gemini 3 Pro It's using that by

default right same thing with GPT.

I can only look at GPT 5.2 But GPT 5.2 codecs this one has

a context of 400,000

right and that's like what so it's 2x So that's one thing

that I'm gonna kind of dock against quad code is I don't

necessarily know why they're doing that I'm still going to

approve the model I'm going to be using this but why is it

that they're not offering us this model with the 1 million

in context window with Our plan that's just a little bit

weird, right?

So let's go back over here and it looks like these are

still running.

We have the speed test going up This one is all fixed and

this is very very fast great work I now need you to launch

sub agents that will implement the fixes without breaking

functionality or writing AI slop code So these are all

working.

Let's launch this one.

Let's see how this one is doing.

So it does have a plan here So what did it say?

So what changes does it want to make so it says change URL?

Which one how is it gonna change URL output a table of your

plan?

Show me the URL before and then show me what you're going

to change it to So let's go here and let's just kind of

make sure that that looks right But that's definitely one

thing to note is that this model selector is a little bit

odd I don't know what you guys think about this,

but definitely something that we want to take note of

right?

Go ahead and implement it.

So I'm going to implement that but even with this context

window drawback I still am noticing some differences and

some improvements and now it does bring Sonnet back into

the equation,

you know with Sonnet models if you guys remember Sonnet 4.5

before we got Opus 4.5 Sonnet 4.5 was by far the best

coding model on the market and then we got Opus 4.5

Everybody forgot about Sonnet 4.5,

but Sonnet 4.6 this is a significant improvement for the

Sonnet suite and it comes at a 40% cost reduction with

about a 25% speed increase So definitely something that you

guys are going to be see me working on and working with in

my vibe coding workflows And with that being said guys,

I'm gonna give it the bridge mind stamp of approval I hope

this helped you guys and you guys are gonna see me using

this in streams And with that being said if you haven't

already liked subscribe and join the discord Make sure you

do so and I will see you guys in the future