Your guide to your life.
An atlas is the world, bound and carried. This one is a child's — every subject, every grade, in their hands, going wherever they go.
Atlas is a complete tutor for children — every subject, every grade, for $5 a month — and it exists because the cost of writing a full curriculum collapsed: a model authors the curriculum at build time, before anyone ships, and never on the child's phone.
Zero to one.
There is a distance between knowing nothing about something and knowing something about it. That first step is the hardest one a person ever takes, and it is the one most likely to be taken alone — by a child with no tutor, in a class of forty, or at a kitchen table with a parent who is also working.
Atlas exists for that step, and it exists to make it visible. Not a grade, not a badge: a map of what a child could not do in September and can do unaided in March, with the evidence attached.
A child should be able to see how far they have come. Almost nothing in education shows them.
Why does a complete education for one price not already exist?
Because a complete education costs too much to write. Not to deliver — to write. Every lesson, every problem, every wrong answer worth anticipating, every hint before the answer, in every grade, in every subject, in a language the child speaks.
It is a fair question to ask, because education is the most-attempted software category there is, and the thing we are describing sounds like the obvious product. The reason is boring and specific, and it is arithmetic. That work is thousands of human-years, and the receipts are public.
| What they spent | What they got | |
|---|---|---|
| onebillion | a decade, nine employees, about £1.6M a year | the best-evidenced early-learning product in the world — two subjects |
| Kitkit School | seven years and a Global Learning XPRIZE | a fixed 529-day sequence, the same for every child |
| Time4Learning | twenty years | complete PreK–12 scope, and the parent decides the order, at $335 a year per student |
| Bloomy | a modern, well-funded start | 1,200+ skills, at $39 a month, per subject |
Nobody chose to be narrow.
Read that table again, because it is not a list of failures. Every one of these is a good product built by serious people. onebillion's results are the ones we would most like to have.
Nobody chose to be narrow. They were priced into it. Scope, depth, adaptivity, price — you could have any two. That was not a strategy problem. It was arithmetic.
What changed.
The authoring cost collapsed. That is the whole event, and it is enough.
A frontier model can now generate a curriculum from a country's own standards, author the lessons and the problems and the hint ladders, anticipate the wrong answers, write the tutor's words, and translate the lot — at build time, on a real machine, once, before anyone ships anything.
It is generated, reviewed, tested and frozen. What reaches your child's phone is already finished.
Does an AI model talk to my child?
No. The model builds the curriculum before the product ships, and no model runs on your child's phone. It is the factory, not the product.
That is a safety decision before it is a cost decision, and there are four reasons. Only the last one is about the phone.
A human reads everything
A lesson generated in the moment has never been read by an author, a teacher, or anyone. Every item Atlas ships has been through a gate before a child sees it.
You cannot measure against a moving target
If your child gets a different question every time, mastery means nothing — there is no stable thing to be right or wrong about. A mastery estimate requires that the item be the same item.
Unlimited questions were already free
One template, a range of numbers and a seed produces as many problems as anyone could ever need, at no extra cost and no extra bytes. Generating them live buys the thing that was already free, and pays for it with verifiability.
And a factory cannot fit in a pocket
A model that must run in order to teach has to load, stay warm and stay correct on a cheap phone with no signal. It cannot.
So what ships.
A deterministic engine. A few kilobytes of code that read your child's mastery state, walk a map of what depends on what, and pick the next thing. Same input, same output, every time. It cannot invent, it cannot drift, and it cannot hallucinate at a seven-year-old.
The intelligence is spent before your child ever opens the app. Which is why it can be given away for the price of a coffee, and why it works with the radio off.
That is the answer to why $5 for everything, and why now. Not a discount. Not a subsidy. The cost structure of education content changed, and nobody had rebuilt the product around it.
One product. One price.
We are building the best tutor in the world, and we are going to charge everyone the same $5 for it.
Not a cheap version for families who cannot pay and a real one for families who can. One product. A child in Los Angeles and a child in Loreto get the same engine, the same curriculum and the same record of what they learned.
Language models build almost everything Atlas ships. They just do not get to improvise at a child.
The best education shouldn't depend on where a child happens to be born. Every design decision in this product exists to make that sentence literally true rather than aspirational.
The best education shouldn't depend on where a child happens to be born.
One email when Atlas opens. Nothing else.