The AI Lab · Silicon Tipster League
Stablebet Model: horse racing tips, tested
Stablebet Model · StablebetOn the starting line
This is our own horse — the in-house StableBet racing model, entered in its own league. In every UK and Irish race it backs the runner it rates highest, logged before the off and settled at SP, so you can see how a purpose-built racing model stacks up against five general-purpose chatbots.
Research, not tips. Every pick logged before the off and settled at industry Starting Price, wins and losses alike. 18+ · please gamble responsibly.
Stablebet Model's scorecard
Collecting — first picks lock at the next meetingNothing settled yet. Stablebet Modelis lined up and its first calls go on the record at the next race meeting. That blank scorecard is the honest starting point — bookmark the page and watch it fill, race by race.
Profit (£1 stakes)
—
ROI
—
Strike rate
—
Bets settled
0
Blind
—Never sees the odds — one £1 pick per race.
- Bets
- 0
- Win%
- —
- Staked
- —
No informed arm
Our model is deliberately built without the market price as an input — its value is an independent read of the race. Showing it the odds would collapse the comparison, so it competes blind-only.
Went its own way: 3 races where Stablebet Model was the lone dissenter while the rest of the field agreed on a different horse.
Stablebet Model's running profit
One £1 pick per race, settled to Starting Price — the whole record, wins and losses alike.
The profit graph draws itself, live.
Each AI's running profit at £1 a bet appears here the moment the first races settle.
Stablebet Model's recent calls
The actual picks and one-line reasons Stablebet Modellogged before each off, newest first. Read the logic and judge it for yourself — the board settles the argument at Starting Price.
| Race | Arm | Pick | Conf. | Reason | Result |
|---|---|---|---|---|---|
| Epsom Downs20:52 | Blind | Sail On Sailor | Medium | Eight runs with a recent win at a rating of 63 give the model solid ground to work from. The model's estimate of 27% sit | pending |
| Newbury20:37 | Blind | Kakirra | Medium | Winner last time from six runs at a mark of 61. The model's estimate of 21% sits below the market's 27%, suggesting the | pending |
| Epsom Downs20:22 | Blind | Happy Banner | Medium | Two placed efforts from thirteen runs, most recently second, with a rating of 73. The model's estimate of 20% sits notic | pending |
| Newbury20:02 | Blind | Raintown | Medium | A veteran of 37 runs with a second last time and one win each from five attempts at Newbury and ten at the trip, rated 6 | pending |
| Epsom Downs19:50 | Blind | Rage Of Thunder | Low | Seventeen runs with a recent win and two victories from ten attempts at this trip suggest a capable performer at this gr | pending |
| Newbury19:27 | Blind | Storm Point | Low | Won its last two starts, including both previous efforts at this course and trip, from just six runs. The model's estima | pending |
| Epsom Downs19:15 | Blind | Jersey Maverick | Medium | Seventeen runs with a recent second and one win from two at this track suggest some affinity for Epsom. The model's 26% | pending |
| Newbury18:52 | Blind | Roman Spring | Low | Thirty-eight runs on the card with two wins from twenty-nine at this trip, most recently third. The model's 11 per cent | pending |
| Epsom Downs18:40 | Blind | Eabha | Low | Two runs on the record, most recently third, with a rating of 81 to show for it. The model sees this at 19 per cent, whi | pending |
| Newbury18:17 | Blind | My Boo Boo | Low | One run, finishing second, which gives the form model some foundation to work with. The model's 10% estimate sits below | pending |
| Epsom Downs18:05 | Blind | Darkest Red | Medium | Nine runs on the record with a second last time, rated 62. The model sees it at 26 per cent while the market's 40 per ce | pending |
| Newbury17:45 | Blind | Ouragan | Low | One run, finishing fifth, gives limited data for the model to assess. The model's estimate of 12 per cent sits above the | pending |
About Stablebet Model
The StableBet model is the machine-learning system behind our AI Race Predictor: a gradient-boosted ensemble trained on years of UK and Irish results, reading each runner's own form, the race conditions and the weather. Unlike the chatbots beside it on this board, it was built for exactly one job — estimating each horse's chance of winning — and it does that with genuinely well-calibrated probabilities: when it says 25%, that horse wins about a quarter of the time.
Here is the honest part, and it is the reason this page exists: well-calibrated is not the same as profitable. The betting market is a formidably sharp forecaster, and our own long-running track record shows the model picking winners at a healthy rate while still losing money at SP, because the prices it takes are slightly worse than the truth it knows. We publish that record in full rather than hiding it.
Putting the model in the league answers a question the chatbots can't: is a purpose-built specialist actually better at this than a general-purpose AI asked for a tip? Same racecards, same one-pick-per-race rules, same £1 at SP. The board settles it in public, one race at a time.
How Stablebet Model reads a race
The model doesn't read a racecard the way the chatbots do — it never sees prose at all. Each runner arrives as numbers: career form figures, official rating against the field, course and distance record, the going, the class, the weather. Out comes a win probability for every horse, and the League simply takes its top-rated runner in each race. The one-line reason you see beside each pick is the model's own published form read for that runner, so the case for the pick and the pick itself always come from the same place. What we get to watch is the specialist's discipline against the generalists' fluency: no story-telling, no hedging, the same cold arithmetic on every card.
Why it runs blind-only
The model competes in the blind arm only, and that is by design rather than a handicap. It has never been given the market price as an input — its whole value as a research tool is that it forms an independent view of a race from the form alone, which we can then compare against what the market thinks. Showing it the odds would collapse the very thing that makes the comparison interesting (and our own testing found that blending in the market price just turns the model into an echo of the bookmakers' board). So its picks here are always the unassisted read: the same 'never sees the odds' test the chatbots face in their blind arm, applied to the one competitor that was actually built for the job.
What to watch on Stablebet Model's board
- Whether the purpose-built specialist beats the general-purpose chatbots over a real sample — the question this entry exists to answer
- How often the model's top pick is simply the favourite, and how it fares in the races where it goes against the market
- Its strike rate versus its profit: the model picks winners at a solid rate and can STILL lose at SP — the gap is the bookmaker's margin at work
- How it compares with The Favourite's baseline row — a model has to out-run the zero-effort strategy before anything else
The rest of the field
Stablebet Model is one of 7 on the board. See how the others are reading the same races:
Questions about Stablebet Model tips
What is the StableBet model backing in the league?
Its top-rated runner in every UK and Irish race on the day's card — one pick per race at £1 level stakes, logged before the off and settled at industry Starting Price, under exactly the same rules as the five chatbots. The pick is whichever horse the model gives the highest win probability, taken automatically with no human override.
Is your model better at tipping than ChatGPT?
That is precisely what this board is measuring, live and in public. The model has a real, published advantage in calibration — its probabilities mean what they say — but calibration is not profit, and a chatbot can luck into a hot streak over a short sample. Watch the settled records side by side rather than taking our word for it.
Why doesn't the model have an informed arm like the chatbots?
Because it was deliberately built without the market price as an input. Its research value is an independent second opinion on each race, formed from form and conditions alone — our testing showed that feeding it the odds just makes it mirror the market rather than read the race. So its league picks are always the blind, unassisted view.
Can I make money following the model's picks?
No — and our own track record page says so in plain numbers. The model wins races at a consistent rate and still loses money at SP over time, because the bookmaker's margin is bigger than its edge. We enter it here as research and entertainment, not as a tipping service. 18+, and never bet more than you can afford to lose.
This page sits inside the AI Lab, where we test whether any betting system makes money (across 26,000+ races, none of them do), and ask the bigger question in does following an AI tipster work?
Gamble responsibly.This page is research and entertainment, not betting advice. No AI here beats the bookmaker's margin, and nothing on it is a signal to stake. Betting should never be a way to make money. If it is affecting you or someone you know, free and confidential support is at BeGambleAware.org. 18+.