St Leger and Irish Champions 2026: The AI Tipster Standings

Eight entrants went to two festivals in the same week. Two came home ahead.

St Leger week ran four days at Doncaster, 10 to 13 September. The Irish Champions Festival sat inside it on the Saturday and Sunday, at Leopardstown and then the Curragh. Between the two meetings the Silicon Tipster League logged 285 valid bets over 24 of Doncaster's 31 races, and a further 98 over 8 at the Curragh. Add the two together and it is 383 bets from 32 races, run by all eight competitors: five frontier models, each entering twice, plus our own two engines and a baseline that just backs the morning favourite.

One afternoon is left out of these figures. Every League pick at Doncaster on Saturday 12 September, St Leger day, was logged after its race had been run, because that morning's collection failed and was re-run later. A pick made after the off breaks the League's one rule, so those 91 picks are excluded here, as they are on the season leaderboard.

The Favourite is sitting on -8.6% across a 2,463-bet all-time record, and every one of the eight entrants is currently behind on its own full record. The League as a whole has now settled 30,049 bets over 76 days. That is the frame this page sits inside. Nothing below changes it. Four good days do not undo a much longer one, and the League exists to show what following an AI tipster actually costs, not to find one that works.

What follows is the two festivals side by side, what they add up to, and then a proper check of whether the market's favourites really had the week the headline numbers suggest, or whether one afternoon did the work.

Doncaster: a good week for reading the card, a thin one for the value arms

Across the three days the League has a valid record for, the field returned −37.1%, and four of the thirteen entrant-arms finished ahead. The best of them was not the favourite. ChatGPT's blind arm, reading the card with no prices attached, led the whole field at +50.0%, six winners from 22 picks. The Favourite was close behind it, backing eleven winners from 23 at level stakes. Gemini finished just ahead on both of its arms, its value arm with three winners from 23 for a small profit, which made it the only informed arm to end the meeting in front on a card where the value arms otherwise lost.

The meeting's own showpiece is the race the League cannot speak to. Christmas Day started 7/4 favourite, chasing the Derby-St Leger double no horse has done since Reference Point in 1987, and finished outside the first four. Highwayman won the St Leger itself at 11/1 for Joseph O'Brien, with Saffie Osborne becoming the first woman to ride the winner of a British Classic. Three of the League's value arms named Highwayman that day, but only after the race had been run, so those picks are excluded and the League has no valid call on the St Leger at all.

Claude's value arm had the opposite week, beaten on all 22 of its valid picks, and our own Stablebet Edge did not find a winner either. The full running of all four days, session by session, is in our St Leger Festival results hub.

The Curragh: the sharpest split of the week, and a Saturday the League never saw

One card, eight races, and the clearest gap between the market and everyone else all week. The Favourite went six from eight, a 75.0% strike rate worth +79.8%. Stablebet Model, our own engine backing the runner it rates most likely to win, found four winners from seven for a small profit of its own. Everything else lost, and five arms lost every single bet: Gemini, Claude, Grok and DeepSeek's value arms, plus our own Stablebet Edge, went a combined 0 from 38.

Four of those blank value arms backed the same beaten horse in the Irish St Leger, which is the specific race our Curragh Group 1 preview, settled already covers in full: Scandinavia won readily as the well-backed favourite, completing a Doncaster-and-Irish St Leger double started twelve months earlier, while eight of the League's ten picks across the four Curragh Group 1s went the same way, onto his stablemate Jan Brueghel, who ran third.

That is the Curragh's closing Sunday. Saturday's card, at Leopardstown, is not in the table above, and it is worth saying plainly why: a scheduling change on 7 September had silently switched off Irish racing across our whole pipeline, and nobody knew until it was fixed on 13 September, the day before this very Curragh card. Leopardstown's meeting fell inside that five-day gap, so no AI picks were logged there at all. That is a fixed gap in our own pipeline, not the League declining to have a view. It is also where the biggest single result of the whole week happened: Hotazhell won the Royal Bahrain Irish Champion Stakes at 50/1, beating the 11/8 favourite Constitution River into third. None of the eight entrants had a view on that race, because none of them were ever asked. The full Saturday and Sunday running is in our Irish Champions Festival results hub.

What the two festivals add up to

Put Doncaster and the Curragh together and the League placed 383 valid bets across 32 races, and lost money at both: −37.1% at Doncaster, −53.8% at the Curragh. Eleven of the thirteen entrant-arms finished behind once both meetings are counted rather than one.

Only two stayed ahead through both. The Favourite was ahead at Doncaster and ahead again at the Curragh, and stays ahead once the two are added together. ChatGPT's blind arm was ahead at Doncaster and gave a little of it back at the Curragh, and also finishes the fortnight ahead overall. Gemini's two small Doncaster profits did not survive the Curragh. Everything else, including both of our own engines and every one of ChatGPT's, Gemini's, Claude's, Grok's and DeepSeek's value-betting arms, lost money once the two festivals are added together. Two arms had no winners at all across both meetings: Claude's value arm from 30 picks, and our own Stablebet Edge from 28.

None of that is a system. It is two entrant-arms clearing a small sample of 32 races, out of thirteen that tried.

The favourites question, checked day by day

Both meetings closed with a short-priced winner. Scandinavia went off odds-on in the Irish St Leger and won it readily. Highwayman did not, as covered earlier, but the rest of the week ran hot for the market leader, and a strike rate like that on one Sunday is exactly the kind of number that deserves checking rather than repeating: whether one bumper afternoon is carrying the whole story, or the pattern held across four days and two countries.

It held on every day the League has a valid record for. Day by day at Doncaster, the Favourite went five winners from eight on the Thursday, three from seven on the Friday and three from eight on the Sunday: never fewer than three winners in a card of seven or eight, never more than five, no single day collapsing and no single day running away with it either. Saturday, the St Leger's own card, is not in that count, because its picks were logged after the races had been run. Add the Curragh's six from eight on top and the week's tally is eleven winners from 23 at Doncaster and six from eight at the Curragh, against an all-time strike rate for the same bet of 29%. Every one of those days beat the long-run figure. That is a real pattern across the week rather than one lucky Sunday, though the Curragh was still the sharpest day inside it.

Which is exactly why the meeting's biggest race sitting inside all that form is worth naming rather than smoothing over. The St Leger was won by an 11/1 shot in a week when the market was mostly right. A good week for favourites in general is not the same claim as a good week for the one that mattered most, and the one that mattered most did not hold.

None of this is an edge. The Favourite returned +37.1% at Doncaster and +79.8% at the Curragh, against -8.6% across its full 2,463-bet record. A good week is what a losing strategy produces some of the time. If it never had good weeks, nobody would follow it.

St Leger and Irish Champions tipster league FAQ

Did any AI actually beat the bookmakers across the two festivals?

Two of the thirteen entrant-arms finished ahead over the 32 races the League has a valid record for, and one of those two is the baseline that just backs the market's own favourite. That is well inside what chance produces over such a small sample, and the wider field lost money. Nothing here is evidence that an AI can beat a betting market.

Why does the same model appear twice in the tables?

Every frontier model runs two separate tests. One reads the racecard with no prices attached and names a winner. The other is shown the market and asked where it disagrees with it. They behave very differently, which is the point of running both, and they are settled and reported separately.

What happened to the Irish Champion Stakes?

It was run at Leopardstown on the Saturday. A scheduling change on 7 September had silently switched off Irish racing across the whole pipeline for five days, 8 to 12 September, fixed the day after, and Leopardstown's card fell inside that window, so no AI picks were ever logged for it. Hotazhell won it at 50/1. The League simply has no record of that race, which is why this page says so rather than leaving a silent gap in the table.

Did your own model have a good week?

Stablebet Model, which backs its own top-rated runner, was modestly ahead at the Curragh and behind at Doncaster. Stablebet Edge, which chases the widest gap between our model's estimate and the market price, lost money at both meetings without a single winner at either. Both engines sit behind on their full season record, which is the number that actually matters.

Will you keep doing this for every festival?

Yes, and each meeting's figures are frozen into a permanent record rather than living only in an article, so they can be compared meeting to meeting on the same terms. The schema, the conventions and what it refuses to claim are written up in How We Measure a Festival.

Bet responsibly. Nothing on this page is a tip or betting advice. Every pick described here was logged before the race and settled at starting price, win or lose. If you bet, stake only what you can afford to lose, and if it stops being fun, stop. Help is at BeGambleAware.org. 18+.