Every driver and every car, rated after every Grand Prix since 1950. A driver’s rating is his own part plus his car’s part. Ratings are on the chess scale: 1000 is an average driver in an average car of that season. A gap of 200 Elo points means about three wins in four (Elo points measure rating gaps; they are not championship points).
Try this: pick Driver + car, then Driver only, then Car only. How much of Antonelli’s 2026 is him, and how much is the Mercedes?
Driver + car is what the model uses to predict a race. Driver only imagines everyone in an average car; Car only imagines an average driver in each car. The Car share view shows two different percentages. The car share (red) is a result: how much of the differences between drivers comes from their cars. The car weight (blue) is a rule of the model: how it splits the Elo points a team earns between the car and its two drivers.
1000 is the average of the cars and drivers that raced that season, so a 1960s rating and a 2020s rating are each measured against their own field (and a race with a tiny field, like Indianapolis 2005 with six cars, does not shift the scale). A line turns pale across races a driver or team did not start. Only finishers are ranked: a retirement, like a missed start, never changes a rating.
The page opens on the hybrid era with its top 8 drivers. Every period button shows the top 8 drivers (the top 6 cars in the Car only view), ranked by the average rating over the period, or by the latest rating for 2025–26. In a rules era, All drivers and All cars show everyone with at least 10 Grands Prix in that era; with more than 12 drivers, teammates share a colour and the second one is dashed. Here 1000 is the average of everyone who raced this season; the article uses the drivers of the latest race, so its driver parts can differ by about 20 Elo points.
It works like the chess Elo rating. Each race counts as a match between every pair of finishers. Beat a better-rated rival and you gain more. The gap to your teammate is yours; the result you share with him is split between the car and the two drivers.
Six parameters, tuned on 1983–2019 with every rules era counting equally: (1) how big each rating jump is; (2) how fast the car weight learns; (3) how hard ratings are pulled back to average when the grid reshuffles; (4) how a rating gap turns into a win chance; (5) a limit on how much one bad race can cost; (6) faster learning after new rules.
The fraction of the shared result that goes to the car (the car weight) is itself learned race by race. It learns from the races where the split makes a difference: drivers in their first season at a new team, and everyone in the first season of new technical rules (rookies are left out in their first season). Each result is compared with how the experienced drivers did in the same race. So a general trend, like veterans beating newcomers, does not push the weight.
Ratings carry over from one season to the next without a jump. In the first five races of each season the model checks how much the order of the teams has shuffled since the winter. The more it has changed, the more all ratings are pulled towards the average. After new technical rules, cars also learn faster for the first races (fading over five Grands Prix). Ratings move only when races happen.
Newcomers, drivers and teams, start 350 Elo points below the average of their first race and move twice as fast for their first 30 Grands Prix (as in FIDE’s rule for new chess players).
Only finishing order: no lap times, qualifying, tyres or weather. The split rests on teammates, so a team built around one driver inflates him and deflates the other. A gap between two ratings means what it means in chess. Levels across eras do not compare: 1000 is always that season’s average. Rookies and new teams start 350 Elo points below the average, a fixed choice made after seeing the results: starting them at zero put newcomers ever further below the field as the decades went by (Cadillac in 2026 ended up more than a thousand Elo points behind Aston Martin).
Article: BIG IF TRUE #003, Who is winning F1, Kimi Antonelli or his Mercedes? (English and Italian). Code and data: github.com/bigiftruebits/bit-003-f1-elo. Ratings after the Singapore GP (round 17 of 23, 11 Oct 2026); this page is not updated after publication.