SAME CAR. WHO'S FASTER?

A driver's finishing position mostly measures their car. Put anyone in the dominant chassis and they look like a champion. So this rating never compares a driver to the field — only to the one person who had the same machinery on the same day: their teammate.

Every weekend produces teammate duels, from both the race and qualifying. Qualifying matters because it is the cleanest comparison that exists: same car, same track, same session, no strategy and no first-lap chaos. Ratings then travel across the grid because drivers switch teams, so two drivers who never raced alongside each other are still connected through the teammates they shared.

Mechanical retirements are discarded. Losing to your teammate on the day your gearbox failed measures the gearbox. Crashes and collisions do count — those are the driver's.

Every driver's strength is fitted jointly, from the whole duel record at once (a Bradley-Terry maximum-likelihood fit), rather than by walking the results in order and nudging a number after each one. That matters: the sequential approach makes the answer depend on where you start and how hard you nudge, and on this exact data it buried a seven-time champion in mid-table purely because a long career kept getting regressed toward the average.

It was tested, not just built. Fitting on every duel before a given season and then predicting that season — which the model never saw — it calls 65% of teammate duels correctly across 2,107 held-out duels, against 50% for a coin flip and 64.3% for simply backing the driver with the better career record. That is a real edge, and a modest one. Two drivers in the same car are genuinely close, and anything claiming to call these duels nine times out of ten is selling something.

The same test decided how much history should count. Weighting recent races more heavily — a duel loses half its weight every 40 races — and counting a race duel as half a qualifying duel predicted better than treating a decade as one lump, so that is what the 2026 table uses. The 2014 version of a driver is not the 2026 one, and a driver's current season shows up within weeks rather than a year late.

The margin of error is part of the answer. The teammate graph is thin — most drivers have only had a handful of teammates — so several of them genuinely cannot be separated. Each rating is published with the range the evidence actually supports, and when two ranges overlap this page says "too close to call" instead of inventing a winner.

The honest limits, because a number worth sharing should carry them:

• It cannot see team orders, split strategies, or who got the upgrade first.
• A driver whose teammate was exceptional will rate lower than their reputation. That is the point — but it is still a comparison of two people, not a measurement of one.
• Race and qualifying duels from the same weekend are correlated, so the ranges are, if anything, slightly optimistic.
• Drivers with fewer than 60 duels are left out of the all-time table, and marked provisional on the current grid (at least 20 needed to appear). One strong season is not a career.
• The rating counts who was ahead, not by how much. The quali gap column shows the margin (median % of lap time to the teammate this season) as evidence — we tested using the margin instead, and it did not predict future duels any better.

Built on the public F1 results archive (Jolpica/Ergast), 2014–2026.