Free AI predictions for AFL & NRL

Alphr ships live predictions across two codes, AFL (head-to-head, margin band, totals) and NRL (head-to-head, margin band, line, totals), both for the 2026 season. Each sport runs its own XGBoost ensemble trained on more than a decade of match data with a strict train / validation / blind-test split. NBA is in active build for the 2026–27 season. Thoroughbred racing is covered by our sister platform, RaceAlpha (racealpha.com), which publishes AI-rated daily selections across every TAB meeting.

Edge is calculated as Model Probability minus Bookmaker Implied Probability (1 / odds). A positive edge means the model rates the outcome more likely than the price suggests, in plain terms, the market is mispricing the runner, side or total. Alphr's filter only releases a tip when that gap clears 7%, which strips out coin-flip plays and concentrates the published board on selections with measurable analytical value. The same filter is applied uniformly across every market and every sport.

Alphr and RaceAlpha publish data, ratings and predictions. They do not take bets, hold customer funds, or operate any wagering account. We carry no commercial relationship with any bookmaker, broadcaster or sporting body. Readers should wager only with licensed Australian operators and treat every selection on this site as informational. Please gamble responsibly, for help, call 1800 858 858 or visit gamblinghelponline.org.au.

Every Alphr sport model follows the same discipline: training data is capped at the end of a defined season, the following season is set aside purely for validation and early-stopping, and the most recent completed season is held back entirely as a blind out-of-sample test. The AFL head-to-head model returned 67.2% accuracy across all 216 matches of the 2025 season (62.8% on edge-filtered selections), with a Pure Alpha variant trained without odds features reaching 74.5%. Every settled tip is logged on the public accuracy page so the track record is independently verifiable, not self-reported.