What the Model Returns
One probability to win, one to place, and the model's order, for every runner in the race.
Win Probability per Runner
Every runner gets a winProb, and the probabilities across a race sum to about 1, so a 20% runner and a 10% runner are directly comparable.
Place Probability over the Places Paid
placeProb follows the places the race actually pays: top 3 with 8 or more runners, top 2 with 5–7, and win only with 4 or fewer.
The Model's Order
modelRank puts every runner in the model's order, 1 being its top pick, and each response names the model version that produced it.
The Same Parameters as Race Form
Predictions take the date, track and race number you already use for race form, so any race you fetch can be predicted with the identical query string.
One Request, the Whole Race Rated
Pass the date, track and race number and get every runner back in the model's order. The model weighs, among other things:
- Recent form and career record
- Performance at the track, distance and going
- Class, and how the runner has performed at that level
- Pace, running style and the likely speed map
- Barrier or box draw, and weight carried
- Fitness since a spell
- Jockey or driver, and trainer form
{
"track": "Randwick",
"raceNumber": 8,
"predictions": [
{
"number": 4,
"name": "Star Runner",
"winProb": 0.2834,
"placeProb": 0.7215,
"modelRank": 1
},
{
"number": 7,
"name": "Fast Track",
"winProb": 0.1923,
"placeProb": 0.5841,
"modelRank": 2
}
]
}Trimmed for length. Runner names are illustrative. The full response covers every runner and names the model version.
How to Read Horse Racing Predictions
Probabilities, not tips
A runner given 28% in a 12-runner race is rated well above the roughly 8% an average runner would get, but it still loses about 7 times in 10. The model tells you how the chances are spread across a race, not who will win it.
Compare within a race
Because the win probabilities sum to about 1, the useful reading is relative: how far ahead the top pick is, how many runners the model thinks are live, and where it disagrees with how the race is being framed.
Beta, and improving
The model is actively being refined, including its calibration, so the percentages can be flatter than they should be. The ranking within a race is the stronger signal today. A public scorecard of how often the top pick wins and places will follow once there is enough recorded history.
What People Build with the Predictions
Set the model against the market
With Premium's live prices, a decimal price of 5.0 implies a 20% chance. Put each runner's winProb beside its price and see where the model and the market part ways.
Horse racing odds API →A baseline for your own model
Use the predictions as a feature or a benchmark alongside race form, career stats, speed maps and class profiles, all from the same API key.
Predictions reference →Ask an AI assistant
The hosted MCP server has a get_predictions tool, so with a Pro key Claude can answer "what are the model's chances in race 3 at Randwick?" directly.
FormFav for AI agents →Coverage
Predictions cover thoroughbred, harness and greyhound racing, produced ahead of each day's races, and the models keep improving as they learn from more results. A runner with little or no race history, such as a first starter, gives the model less to work with.
Horse Racing Predictions API FAQ
Predictions from $10 a Month
Pro includes predictions, results, speed maps, form badges and jockey, trainer and track bias statistics, and starts with a 7-day free trial. Or begin with a free key for race form and career stats.
Predictions are probabilities, not advice, and Pro is for personal use. If gambling is causing you harm, support services are listed in our Terms.