Built for people who read the race before the price.
FormRace puts machine-learned model ratings next to live market money, so the races worth your attention are obvious and the rest of the card is not your problem.
How It Works
Data Ingestion
We pull race fields, form, results, and live odds from every Australian thoroughbred meeting — updated continuously throughout the day.
Model Scoring
The default ML model uses 94 inputs per runner — market price, pace, class, barrier, weight, ratings and race context — to produce a calibrated win probability.
Market Comparison
Predicted probabilities sit beside live market odds so you can investigate disagreements. A profitable betting edge has not been established.
Platform Features
AI Race Intelligence
Machine-learning win probabilities for every runner in every race, updated as new data arrives.
Market Intelligence
Live odds tracking, market movers, and steam/drift detection across major market sources.
Form Analysis
Deep form profiles for horses, jockeys, and trainers with historical performance breakdowns.
Value Detection
Automated identification of model overlays where our model price sits below the market price.
Under the Hood
The default XGBoost ranker and classifier use 94 inputs per runner across 20 groups — market price, pace projection, weight, barrier, class, ratings, jockey and trainer signals, and race context — then calibrates and blends with the public market to produce a win probability for every runner at every Australian flat meeting. Across 15,048 walk-forward races it picked the winner 31.1% of the time against 34.7% for the market favourite on the same races.
31.1%
Model Top Pick
Nov 2025 – Jul 2026
34.7%
Market Favourite
Same races
15,048
Races Tested
Walk-forward
We are 3.6 points behind the market favourite and we publish it. View full model track record →
Frequently Asked Questions
What is FormRace and how does it work?
FormRace is a race intelligence platform for Australian thoroughbred flat racing. The default served model uses 94 inputs per runner, including market price, pace, weight, barrier, class, ratings, jockey and trainer signals, and race context. It produces a calibrated win probability beside the market price; results are published after the race.
How accurate are FormRace's model ratings?
The historical 15,048-race test (Nov 2025 – Jul 2026) shows 31.1% top-pick wins versus 34.7% for the starting-price favourite and -17.1% flat-stake ROI. A missing-odds fallback in that test requires a clean point-in-time rerun; see the full month-by-month record and caveat on the model performance page.
Does FormRace beat the market?
No, and we do not claim to. Public betting markets are the strongest single predictor in racing and our own published measurements say so. What FormRace does is compress the reading — every meeting, every runner, 94 features each — into a board you can scan in minutes, with 4 checks on whether the favourite is vulnerable and a public record of what it said beforehand.
Is FormRace free to use?
Yes. The free tier gives you the daily race intelligence briefing, public proof, and selected public race-analysis workflows — no credit card required. Pro (A$49/month) unlocks the full AU thoroughbred intelligence board: all meetings, full model ratings, the model's own reasoning, market intelligence, PULSE context, and track-bias intelligence.
What makes FormRace different from Punters.com.au or Racenet?
FormRace publishes its model's strike rate against the market favourite, month by month, on the same races — including the periods where the model is behind. It also shows the reasoning behind each rating: sectional evidence, pace projection, stewards' notes, and the specific checks that make a favourite look vulnerable. Most racing platforms publish selections or star ratings without either the methodology or the record.
What data sources does FormRace use?
FormRace ingests data from Racing Australia for race fields, form, and results. Live odds are pulled from major Australian market sources including TAB, Sportsbet, and Ladbrokes. The ML model is trained on historical race data and updated continuously as new form data becomes available.