Lotto · Eurojackpot · Vikinglotto · iPhone
A draw machine in your pocket.
People pick lottery numbers by preference: birthdays, lucky numbers, familiar patterns on the slip. The draw has none. LottoAIPredictor builds a simulated draw machine from the full draw history and gives you rows the way they are actually drawn, not the way people choose them. Everything happens on your iPhone.
- Free download
- Less than one row a month
- Gaming data never leaves the phone
- Age limit 18
The drum on this page is an illustration. In the app, the simulated draw machine draws real row suggestions the same way, ball by ball.
How it works
Three steps from installation to a filled-in ticket. No account, no login.
The history is fetched, the draw machine is built
The full draw history is downloaded to the phone — Lotto back to 2012, Eurojackpot and Vikinglotto to 2022. From them, a simulated draw machine is built per game — locally, in seconds — and rebuilt for every new draw that comes in.
You set the frame
Choose how many rows you want, and combine the six models any way you like, one or all at once. Every row is labelled with the model that drew it. Add your own rules: numbers to leave out, a similarity limit against past draws, balance between high and low. If you want to stay away from the numbers most people play, that's one switch.
The rows are read aloud while you type
Put the rows in the basket, switch to the operator's app, and the numbers are read out calmly while you type — with the number visible in the Dynamic Island. The tickets are checked automatically against the right draw afterwards.
Digital lottery draw machine
Advanced mathematical and statistical methods give you rows that match the real draw machines.
The draw machine in the app
The full draw history is downloaded to your phone: Lotto from 2012, Eurojackpot and Vikinglotto from 2022. From it, a simulated draw machine is built for each game, drawing ball by ball the way the real one has. Every suggestion is a simulated draw with the same spread, sum and mix as the real ones. When a new draw arrives, the machine is rebuilt with it included. Everything happens locally on your iPhone.
Six models, freely combined
- Neural network — trained on the full history, draws number by number with context from earlier results
- Neural mirrored — the same network, every number swapped for its mirror image
- Pattern analysis — Bayesian frequency, pair correlations, gap hazard and Markov transitions
- Smart Combo — an ensemble merging several methods into one suggestion per row
- Frequency model — statistically weighted, works without training from the very first draw
- Random + variance — pure chance with at most a 40 % pull towards the model's profile
Pick one or all six at once. Every row is labelled with the model that drew it, so you can follow the hit rate per model over time.
Your rules, every model
- Exclusion — numbers to leave out, picked straight from the analysis
- Similarity limit — the maximum number of shared numbers with any earlier draw, default 3
- Balance — the split between high and low, odd and even
- Delivery guarantee — rows that break the rules are repaired rather than discarded, so the quantity you ordered is delivered
For every choice the app shows the probability as a decimal: that your assumptions hold, and of a prize given that they do. The more assumptions, the smaller the outcome space, from millions of combinations down towards a tombola.
Avoid popular rows
Prizes are shared, so a row a thousand others hold is worth less than one you hold alone. The official prize statistics for every draw since 2012 show how often each number is played. With one setting, only rows shared with fewer people than an average row get through, for every model. The odds are the same; the prize is shared with fewer.
The measurements in the app
- Distribution match — how closely the suggestions resemble real draws in sum, spread and pattern, typically 95–97 % after a rebuild
- Bias test — whether any balls are drawn more often than others, and how many draws it would take to see it
- The numbers people play — what the country picks, measured from how many shared each prize
- Accuracy dashboard — hit rate per model against pure chance, with a statistical noise band
Three games
Lotto (7 of 34), Eurojackpot (5 of 50 + 2 star numbers) and Vikinglotto (6 of 48 + 1 Viking number). The app follows the country you choose.
The odds are the same
Lotto, Eurojackpot and Vikinglotto are games of chance. Every combination has exactly the same probability, and no app can predict the next draw. It is in the terms, and you confirm it before the app can be used.
The odds
The probability changes only with how many different rows you play, never with which numbers are on them.
The chance of at least one prize, all tiers, given how many different rows you play in one draw:
| Rows | Lotto | Vikinglotto | Eurojackpot |
|---|---|---|---|
| 1 | 0.35 % | 4.7 % | 3.1 % |
| 5 | 1.7 % | 21 % | 15 % |
| 10 | 3.5 % | 38 % | 27 % |
| 22 | 7.5 % | 65 % | 50 % |
| 44 | 14 % | 88 % | 75 % |
| 66 | 21 % | 96 % | 88 % |
| 80 | 25 % | 98 % | 92 % |
| 100 | 30 % | 99 % | 96 % |
1st and 2nd prize with the same number of rows — those are the ones people want to win:
| Rows | Lotto | Vikinglotto | Eurojackpot | |||
|---|---|---|---|---|---|---|
| 7 correct | 6 + bonus | 6 + Viking | 6 correct | 5 + 2 stars | 5 + 1 star | |
| 1 | 0.000019 % | 0.00013 % | 0.0000016 % | 0.0000065 % | 0.00000072 % | 0.000014 % |
| 5 | 0.000093 % | 0.00065 % | 0.0000081 % | 0.000033 % | 0.0000036 % | 0.000072 % |
| 10 | 0.00019 % | 0.0013 % | 0.000016 % | 0.000065 % | 0.0000072 % | 0.00014 % |
| 22 | 0.00041 % | 0.0029 % | 0.000036 % | 0.00014 % | 0.000016 % | 0.00031 % |
| 44 | 0.00082 % | 0.0057 % | 0.000072 % | 0.00029 % | 0.000031 % | 0.00063 % |
| 66 | 0.0012 % | 0.0086 % | 0.00011 % | 0.00043 % | 0.000047 % | 0.00094 % |
| 80 | 0.0015 % | 0.010 % | 0.00013 % | 0.00052 % | 0.000057 % | 0.0011 % |
| 100 | 0.0019 % | 0.013 % | 0.00016 % | 0.00065 % | 0.000072 % | 0.0014 % |
1 − (1 − p)N, where p is the chance per row computed from the rules of the game and the prize tiers. Always as a decimal, never an exponent. The full prize ladder per game is in the terms.
Full control of the outcome space
Your rules apply to every model, and the app always says how much they shrink the outcome space.
Exclusion rules
Tap numbers in the hot/cold and pattern analysis to leave them out — the rules are saved and enforced by every model and tool.
Similarity limit against history
Suggestions are limited to at most 3 (adjustable) numbers in common with any historical draw — historically, almost no winning rows share 4+ numbers.
Delivery guarantee
Rows that break your rules are not discarded — they are repaired with the min-conflicts heuristic (only the conflicting numbers are swapped), so the ordered quantity is delivered even under tight rules.
The odds as a decimal
Always as a decimal ("0.0000014 %"), never an exponent — for the full outcome space and given your assumptions, with the reduction stated as a percentage.
Avoid popular rows
One setting, applied to every model: only rows shared with fewer people than an average row get through. The basis is the official prize statistics for every draw since 2012 — how many won in each tier — and that reveals how often each individual number is played. No one else's ticket is involved: the app only sees how many landed in each prize tier. The app tells you how much your rows are shared. The chance of winning is the same.
Under the bonnet
A neural network is at bottom one big formula with thousands of adjustable numbers — coefficients (also called weights). For Lotto the network looks like this: 109 input signals go in, pass through four layers of 64, 32, 16 and 8 "neurons", and out come 34 numbers — one probability for each ball in the drum. Around 10,100 coefficients in total, all of them learned from real draws. Three such networks are trained independently and averaged.
The Lotto network: 109 signals → 64 → 32 → 16 → 8 neurons → 34 outputs.
What are the 109 signals? For each of the 34 balls the app measures how often it has come up recently (recency-weighted frequency) and whether it is "hot" or "cold". On top of that come the shape of the previous draw (sum, spread, odd/even and the spread across the number range), how far the row has come, and which numbers have already been chosen in the row being built. The statistics the network sees during training are always computed from the draws before the draw it is practising on — never from the answer itself.
How are the coefficients made? They start as small random numbers. Then the network is shown the whole draw history, number by number: it guesses the next ball, sees the answer, and every coefficient is nudged a tiny step in the direction that would have made the guess better. This is called error correction (gradient learning). One Lotto draw gives 8 such exercises (7 main numbers + the bonus number), so one pass over the Lotto history is roughly 6,100 small adjustments. The last five per cent of the draws are kept out of the training, and it stops by itself when the network stops improving on them — in practice after one to three passes, done in seconds, locally on the phone. That is the point: a network allowed to practise at length on a drawing machine with no memory only learns the history's own randomness by heart. In the app you set the upper limit (5–50 passes) and the step length yourself.
This is how it becomes a row: the network proposes probabilities for the next ball, the app makes a weighted draw (not just the top candidate — that gives variety), the chosen number is locked, and the process repeats until the row is full. For every row you have ordered, five such candidates are drawn, ranked by structure, clustering, spread and weighting — with your own sliders — and the best one is delivered. The selection is deliberately narrow: the harder the app sifts, the further the rows drift from a real draw.
Every game has its own network, shaped by the rules of the game. Eurojackpot and Vikinglotto also have a small bonus network for the star and Viking numbers, included in the coefficient counts below:
| Game | Shape of the network | Coefficients |
|---|---|---|
| Lotto | 109 → 64 → 32 → 16 → 8 → 34 | ~10,100 |
| Eurojackpot | 157 → 64 → 32 → 16 → 8 → 50 | ~14,300 |
| Vikinglotto | 151 → 64 → 32 → 16 → 8 → 48 | ~13,300 |
Follow up your games
From row suggestion to result — without typing anything twice.
Ticket history with win rating
Played tickets are stored locally for 6 months with game, draw date and result — every ticket gets 0–5 stars based on the prize tier.
Automatic result check
Draw results are fetched shortly after every draw, and active tickets are checked automatically against the right draw.
Hit rate per model
The hit rate per model is shown against random expectation, with a statistical noise band, in a dedicated accuracy dashboard.
Rows read aloud
Play mode guides you: press play, switch to the operator's app, and the numbers are read out calmly while you type — with Dynamic Island support. Or choose the display for reading them off yourself.
Three games — full history on all of them
The app follows the country you choose. In Norway all three apply; in other countries only the games offered there.
Lotto
SaturdayPick 7 of 34 numbers
Norway's classic. The bonus number is drawn automatically from the same drum as the main numbers — you only pick the seven.
✓ Every draw since 2012, in the app
Eurojackpot
Tuesday and FridayPick 5 of 50 + 2 star numbers of 12
Europe's big jackpot with two draws a week — main numbers and star numbers are drawn from separate drums.
✓ Every draw since March 2022, in the app, when the star numbers became 1–12
Vikinglotto
WednesdayPick 6 of 48 + 1 Viking number of 5
The Nordic joint game — six main numbers plus your own Viking number from a separate drum.
✓ Every draw since January 2022, in the app
Subscription
Free download on the App Store. For the first 21 days everything is available, with adverts. After that a subscription is required — and that is what keeps the app running.
LottoAIPredictor depends on backend infrastructure to work: the draws are fetched, checked and stored on a server that has to be run and paid for, however few or many people use the app. The analysis and the model themselves run locally on your phone, but without the draw history everything stops.
So we would be grateful if you help us keep the service running by taking out a subscription at the App Store minimum price per month. The subscription is bought, managed and cancelled in the App Store, and gives full access without adverts. Thank you for your support.
- All six models
- All the games in your country
- Automatic fetching of draws
- No adverts
- Bought and cancelled in the App Store
After the 21 days
The trial applies once per Apple ID, from the first installation. When it is over, a subscription is required to keep using the app: without one, the app stops at the launch ceremony — the drum empties ball by ball, and "Subscribe" is the way forward.
No partial access, no hidden fees. The subscription renews automatically until it is cancelled in the App Store, and cancellation takes effect from the next renewal. The details are in the terms.
Privacy in brief
The app is built so that your data stays on the phone. The full policy is on the privacy page.
No user account
No registration, no login, no e-mail. We do not know who you are.
Everything local
Rows, tickets, settings and the trained models live only on your iPhone, and are deleted when you delete the app.
One kind of traffic
The app fetches draw results from our result service. The request says which game and from which date — nothing about you.
Adverts only with consent
Adverts are shown only during the trial, and only after you have responded to Google's consent form and Apple's tracking prompt.
Frequently asked questions
Can an AI app predict the lottery numbers?
No. Lotto, Eurojackpot and Vikinglotto are games of chance, and no model can predict winning numbers. The probability per row is the same whatever numbers are on it. What the app does is describe draws that have happened, put your rows together according to the rules you choose, and show you the real probability.
What happens when the 21 days are over?
Then a subscription is required to keep using the app. Without one, the app stops at the launch ceremony: the drum empties ball by ball, and the "Subscribe" button is the way forward. The trial applies once per Apple ID — neither reinstalling, a new phone nor changing the date gives a new one.
What does LottoAIPredictor cost?
The download is free. For the first 21 days you have full functionality, with adverts. After that a subscription at the App Store minimum price per month is required. The app depends on backend infrastructure for the draws, and the subscription keeps it running. The subscription is bought, managed and cancelled in the App Store, and cancellation takes effect from the next renewal.
Does the app send my gaming data anywhere?
No. The rows, tickets, settings and trained models live only on the phone. The only traffic from the app is fetching draw results, and that request contains no personal data. The app has no user account.
What is an AI lottery model?
In LottoAIPredictor there are six generation models: a neural drawing machine trained on historical draws, the same one mirrored (every number swapped for the maximum plus one minus the number), pattern analysis with Bayesian frequency analysis and Markov transitions, Smart Combo which combines several methods, a purely statistical frequency model — and "Random + variance", where chance gets a gentle pull towards the network's profile, never more than 40 % weight.
What does it mean that the rows are "optimised"?
That the rows are built by rules instead of chance: frequency weighting from the history, global de-duplication so you never get two identical rows, and exclusion rules that spread the numbers. Rows that break your rules are not discarded — they are repaired with the min-conflicts heuristic (only the conflicting numbers are swapped), so you always get the quantity you ordered. If you also reduce the outcome space with your own assumptions, the app shows the probability given that the assumptions hold — and says at the same time that the overall probability is unchanged.
Can the network predict draws?
No. During development we measure the network against draws it has never seen, and there it sits within the margin of error of purely random rows. That is also what you should expect, since the drawing machine has no memory. What the app can do is give you rows with the same properties as real draws — not know which one is coming.
How big is the neural network?
Every game has its own network. For Lotto: 109 input signals, four hidden layers of 64, 32, 16 and 8 neurons, and 34 outputs — one probability per ball — around 10,100 learned coefficients in total. Eurojackpot and Vikinglotto have ~14,300 and ~13,300, including the small bonus network for the star and Viking numbers. Three such networks are trained independently and averaged. Everything is trained locally on your iPhone on every draw since the game began. The layers can be switched on and off in the training settings, including a wide entry layer of 128 neurons.
Which games does the app support?
Lotto (7 of 34), Eurojackpot (5 of 50 + 2 star numbers) and Vikinglotto (6 of 48 + 1 Viking number). The app follows the country you choose: in Norway all three, in other countries only the games offered there. Draw results are fetched automatically, and your tickets are checked against the right draw.