New dots three-note preview is absurd. A

new AI just scored 42 out of 42 on one

of the hardest math tests in the world.

That's the International Math Olympiad.

Full marks, every single problem. But

that's not even the wild part. The team

behind it just gave that same AI away

for free. Anyone can use it. And the

reason it matters isn't the math score.

It's what happens after the math score.

This is dots three-note preview. It

comes from a lab called dot studio, part

of the company Xiaohongshu, and it

dropped on August 14th, 2026. Here's why

I want you to pay attention today. Most

AI models get judged on one question.

You ask something, it answers, done. But

dots three-note isn't built for that.

It's built to keep working on something

for a long time. Not seconds, not

minutes, hours, even days. Think about

that for a second. Most AI forgets what

happened 5 minutes ago. This one is

built to remember what happened

yesterday and use it today. So let's

break down what's actually inside this

thing. Hey, if we haven't met already,

I'm the digital avatar of Julian Goldie,

CEO of SEO agency Goldie Agency. Whilst

he's helping clients get more leads and

customers, I'm here to help you get the

latest AI updates. And Julian Goldie

reads every comment, so make sure you

comment below. dots three-note has 280

billion parameters. Parameters are just

the tiny knobs inside an AI brain that

help it think. But here's the trick. It

only turns on 16 billion of those knobs

at a time. So it's huge, but it's not

slow. Picture a giant toolbox with 280

tools in it. You don't carry the whole

box every time. You just grab the 16

tools you actually need for the job in

front of you. That's why this model can

be big and still fast. It also has a

memory window of 512,000 tokens. Tokens

are just chunks of words. So think of it

like a notebook with over 500,000 words

of space. That's enough room to hold

long documents, long conversations, even

hours of video all at once. And it

doesn't just read text. It understands

pictures. It understands video. It

understands sound. One model, four

senses. Let's talk about the actual

breakthrough. It's called Tempo. Here's

the problem Tempo solves. Imagine an AI

doing a task with 300 steps, booking a

trip, running a shop, planning a

wedding. You don't find out if it worked

until the very end. So, how do you know

which of those 300 steps were smart and

which ones were lucky? Old training

methods struggled with this. They only

graded the AI at the finish line. Tempo

breaks the job into smaller chunks.

After a few steps, the AI stops and

checks itself. It asks, "Am I still on

track?" It doesn't just guess, it

reasons it out. It even uses tools to

double-check. Then it keeps going, or it

changes course. Picture a hiker checking

a map every 20 minutes instead of

walking blind for 6 hours and hoping

they end up in the right place. That's

Tempo. And here's the coolest part. When

the AI gets something wrong, it doesn't

just shrug and move on. It writes down

what it learned, literally. It updates a

file called memory.md, like a notebook

it keeps for itself. So, if it makes a

wrong guess about how something works,

it fixes the note and remembers the fix

later. That's not a chatbot answering a

question, that's a system learning while

it works. Now, here's the part that

actually matters for the video today.

25% of the way in, let's pause for a

second. If you're a business owner, this

whole story might feel like it's

happening far away from you. Big

numbers, big labs, big science fair.

But, it's not far away. Every one of

these releases changes what your

business can do this month. That's the

whole reason I built the AI Profit

Boardroom. We take AI news like this and

turn it into a plan you can actually

use. Daily tutorials, weekly live

coaching calls, a 30-day roadmap, and a

way to connect with other business

owners near you who are already using

tools like this to save time and bring

in more customers. This is an

AI-generated video, by the way, made

with an AI avatar. If you want help

turning news like this into real

automation for your business, the link

is in the comments and description.

Okay, back to the model. Let's look at

how they tested this. One test was a

video game called Slay the Spire 2.

Nobody trained the AI on this game ahead

of time. It had to figure out the rules

by playing health, cards, gold, shops,

enemies. It had to plan ahead because

one bad choice early on could ruin the

whole run. It made it to floor 33. Why

does that matter to you? Because that's

exactly what running a business feels

like. You don't get the full rulebook.

You learn as you go and one early

decision can shape everything after it.

Another test use something called A R C

A G I 3. This one drops the AI into a

totally unfamiliar puzzle with no

instructions. It has to figure out the

rules just by trying things. What moves?

What's dangerous? What opens a gate? The

model solved all six levels. It took 320

tries to get there. But here's the thing

that matters. It wasn't repeating

something it memorized. It was learning

the rules of a brand new world one

mistake at a time. Then there's a demo

that anybody watching this can actually

picture. They gave the model a floor

plan, a fridge's measurements, and a

counter top size. Then they asked, "Will

this fridge actually fit?" The AI had to

pull information from a picture, a spec

sheet, and plain text, then reason it

through. It also told them to

double-check the measurement in person

before trusting it completely. That's

useful. That's not a party trick. That's

a model being honest about what it

doesn't know for sure. They also used it

to design a full app for a headset based

on nine picture examples of an app

style. It planned the layout, wrote the

code, built the project, and got it

running. Nearly 2,000 lines of code

planned and built on its own. Now let's

talk about two new tests Dot Studio

built alongside this model because they

explain exactly why long messy tasks

matter more than quick answers. One is

called Vibe Search Bench. It tests

something normal search tools are bad

at. Real people don't always know

exactly what they want. They ask a vague

question, then adjust as they learn

more. This benchmark has 200 tasks built

around that exact kind of back and

forth. And even the best AI models only

scored around 30 out of 100 on it. That

tells you how hard real messy searching

still is for AI. The second one is

called VIBE Life Bench. This one is

wild. It drops an AI into a simulated

life with 200 long tasks spread across

10 everyday situations using 22

different digital services, and the

world keeps changing while the AI is

working, sometimes without telling it.

So, the AI has to notice changes on its

own. It has to remember goals from days

ago. It has to decide when to act and

when to wait. Here's the line that stuck

with me. This benchmark isn't asking if

AI can answer your question. It's asking

if AI can remember what mattered 3 weeks

ago. Let's bring this back to something

real for a second. A model that can

admit when it gets wrong and fix the

guess is doing something most people

struggle with, too. We rush. We guess.

We don't always double-check ourselves

before moving forward. Maybe that's the

human lesson hiding inside a robot

notebook file. Check your work. Write

down what you learned. Keep going

anyway.

And that's exactly the mindset behind

everything we teach inside the AI Profit

Boardroom. New AI tools show up every

single week. Most business owners don't

have time to test them all and figure

out what's actually useful. That's the

whole point of what we built. We take

releases like dots three note and we

turn them into simple step-by-step ways

to save time, bring in more leads, and

get more customers without you needing

to understand any of the technical stuff

yourself. Inside you get weekly coaching

calls where you can ask questions live,

daily tutorials that walk you through

exactly how to use new AI tools like

this one, a 30-day roadmap so you're not

guessing what to do first, and a way to

connect with other business owners near

you who are already putting these tools

to work. This video was made using an AI

avatar, and everything you just heard

came from real research, not guesswork.

If you want the deeper breakdown, the

prompts, and the real playbooks behind

tools like this, join the AI Success

Lab. It's a free community. You'll get

the notes from this video plus access to

over 75,000 members figuring this stuff

out together. Links are in the comments

and description. AI didn't just learn to

answer questions this week, it learned

to check its own work and that changes

what it can actually be trusted to do

for you.