Which AI predicts football best? Models ranked by real bets
We compare AI models on bets recorded before kick-off. For each model we count how many bets won, the average odds, the ROI and the profit at a stake of one unit. Below are the results by market, plus the bets on upcoming matches.
There are 20 AI models in the ranking right now. Since 24 September, 7275 bets have been settled across 472 matches: 55% won, for an overall ROI of −1.9%. For comparison, backing the bookmakers' favourite returned an ROI of +9.8% on the 435 matches of this sample with a known closing line. ChatGPT leads on profit (+21.54 units), but the sample is still small and the order changes from round to round.
Ranking by profit
Ranked by profit in units at a stake of 1; won — excluding voids; ROI — profit per unit staked; pending — bets on matches not yet played.
CLV shows how much higher our odds were than the closing odds before kick-off; closing odds were available for 6235 of 7275 settled bets, and a row displays it once a model reaches 20 bets with closing odds.
The comparison rows are calculated on the same matches; the favourite's odds are the closing odds before kick-off, and matches without them are left out of that row.
There are few matches so far: the order changes from round to round.
AI model results by league
| AI model | EPL | Serie A | Bundesliga | LaLiga | Ligue 1 | Ekstraklasa | Super Lig | UCL | UEL | UECL | Nations League |
|---|---|---|---|---|---|---|---|---|---|---|---|
| ChatGPT | 23/42 | 12/23 | 9/27 | 18/34 | 16/28 | 25/40 | 17/26 | 23/35 | 27/50 | 33/52 | 55/93 |
| 24/43 | 16/24 | 16/28 | 27/34 | 22/28 | 29/42 | 15/25 | 25/32 | 32/51 | 29/49 | 44/62 | |
| Nemotron Super | 20/43 | 15/24 | 12/29 | 24/38 | 20/29 | 22/40 | 15/21 | 23/35 | 29/52 | 29/52 | 49/90 |
| 21/43 | 11/22 | 11/28 | 19/38 | 20/29 | 23/43 | 18/26 | 24/36 | 30/53 | 28/53 | 54/94 | |
| — | — | — | — | — | — | — | — | — | — | 14/20 | |
| Nemotron Lightning | — | — | — | — | — | — | — | — | — | — | 14/23 |
| Mistral | 19/43 | 15/24 | 16/29 | 21/36 | 17/29 | 24/43 | 14/26 | 23/36 | 33/54 | 26/54 | 34/67 |
| 17/37 | 9/18 | 10/20 | 17/33 | 17/27 | 22/32 | 9/16 | 9/13 | 18/31 | 18/34 | 37/72 | |
| Nemotron Ultra | 19/41 | 12/21 | 14/28 | 16/31 | 17/27 | 15/29 | 13/18 | 13/19 | 21/33 | 24/51 | 47/83 |
| DeepSeek Flash | 20/43 | 12/22 | 14/29 | 19/36 | 15/29 | 22/40 | 18/25 | 23/36 | 28/51 | 25/52 | 52/92 |
| MiniMax | 19/43 | 10/24 | 11/27 | 20/38 | 16/29 | 23/41 | 17/25 | 21/35 | 27/53 | 23/50 | 55/92 |
| 18/43 | 12/24 | 15/28 | 19/38 | 18/29 | 23/43 | 15/25 | 23/34 | 34/53 | 21/50 | 53/91 | |
| Kimi | 16/42 | 13/24 | 14/28 | 17/38 | 18/29 | 21/42 | 19/26 | 19/34 | 27/54 | 27/54 | 36/64 |
| DeepSeek | 17/43 | 9/23 | 13/27 | 17/38 | 16/29 | 22/41 | 11/25 | 23/34 | 31/52 | 27/53 | 35/62 |
| Xiaomi MiMo | 21/42 | 15/23 | 14/28 | 20/38 | 18/28 | 23/42 | 15/26 | 27/36 | 20/53 | 26/53 | 51/93 |
| Claude | 21/41 | 9/23 | 13/28 | 13/38 | 15/29 | 22/42 | 15/25 | 28/35 | 28/53 | 31/53 | 53/94 |
| 17/42 | 12/23 | 11/26 | 17/38 | 15/28 | 24/43 | 18/25 | 24/33 | 31/53 | 23/48 | 45/89 | |
| Qwen | 17/42 | 12/23 | 15/29 | 16/37 | 19/29 | 25/42 | 16/26 | 24/36 | 26/53 | 25/52 | 52/94 |
| ERNIE | — | — | — | — | — | — | — | — | — | — | — |
| — | — | — | — | — | — | — | — | — | — | 7/10 |
Each cell shows won out of settled; from 10 settled bets the colour indicates profit: green for profit, red for a loss.
A competition name leads to the AI ranking for that competition.
What AI models bet on and what wins
| Market | Share of bets | Won | Avg odds | ROI |
|---|---|---|---|---|
| Handicap | 24% | 833 of 1580 | 1.94 | −0.9% |
| Total goals | 24% | 935 of 1758 | 1.73 | −9.8% |
| Match result (1, X, 2) | 24% | 970 of 1719 | 2.04 | +10.0% |
| Both teams to score | 20% | 818 of 1444 | 1.62 | −8.8% |
| Double chance | 5% | 240 of 364 | 1.61 | +2.0% |
| Both teams to score — no | 3% | 100 of 195 | 2.03 | +0.2% |
Share of all AI bets, including matches not yet played; won and ROI are based on settled bets.
How we compare AI models
One match card. The models get data on form and line-ups, the league table, head-to-head record and odds. They do not see each other's answers.
One bet per model. Each AI model picks one available market: match result, total goals, handicap, double chance or both teams to score. If we have no odds for the chosen market, the bet does not count.
The same odds for every model. We settle at the average bookmaker odds at the time of collection. The price the model named itself is not used. That way every participant is judged by the same rules.
Ranked by profit. Rankings depend on the odds taken as well as how many bets won. We also show ROI, average odds and results in individual leagues.
Switching models does not reset the record. A participant is one developer's AI. When we move it to a different model, its bets keep counting towards the same total.
We do not publish the models' full answers. The match page shows each AI model's bet and the consensus, the pick of the largest group of models. The models' detailed reasoning is not part of the ranking.
Questions about AI bets
Which AI predicts football best?
ChatGPT leads on profit across all competitions right now: 258 of 450 won, +21.54 units at a stake of 1. It is too early to declare a clear winner: the models have between 21 and 472 settled bets each, and the rankings change from round to round.
Can you bet on AI predictions?
Here is the record so far: all AI models combined returned an ROI of −1.9% across 7275 bets; backing the bookmakers' favourite returned +9.8% on the 435 matches with a known closing line; the AI consensus, our published prediction, returned −2.9% across 472 matches. These are past results, not a promise: betting means risking money, and the sample is still small.
How do the AI models differ from each other?
In style — what they bet on and at what prices: 73% of Nemotron Super's bets are on the match result; 37% of Mistral's bets are on both teams to score; the highest average odds belong to Nemotron Lightning (1.95), the lowest to Sonar (1.57).