Tesla Reports Six Million Miles Between Major Collisions, Not a Controlled Comparison

Tesla says vehicles using Full Self-Driving (Supervised) now travel an average of 6,006,200 miles between major collisions. The October 2026 update improves on the approximately 5.3 million miles reported in February, but the manufacturer’s comparison with U.S. drivers does not place the groups under identical operating conditions.

The figure covers North American driving over a rolling 12-month period and is updated quarterly. Tesla defines a major collision as one involving airbag deployment. It counts qualifying collisions regardless of responsibility and attributes an incident to FSD Supervised when the system was active at any point during the five seconds before impact.

What the six-million-mile figure measures

Miles between collisions is an exposure-based measure: the accumulated distance is divided by the number of qualifying events. A rising interval indicates fewer recorded collisions relative to the reported mileage, so the increase from roughly 5.3 million to just over 6 million miles is a favorable movement within Tesla’s own reporting framework.

Tesla compares that result with 2,031,221 miles between major collisions for manually driven Teslas using active safety features. It also cites an estimated U.S. average of 660,164 miles. Those figures make the FSD interval nearly three times the manual-Tesla result and about nine times the national estimate.

The same ordering appears in Tesla’s minor-collision figures. The company reports 1,760,823 miles between incidents with FSD Supervised engaged, compared with 685,390 miles for manually driven Teslas with active safety features and an estimated national interval of 225,322 miles. Minor collisions are those in which airbags do not deploy. Tesla separately says FSD has six times fewer off-highway collisions than the estimated U.S. average.

Why the comparison is not like-for-like

The national benchmarks are estimates derived from federal crash and mileage information. The underlying government statistics combine multiple reporting systems and cover crashes of different severities; readers can inspect the National Highway Traffic Safety Administration’s annual traffic-safety tables for the structure and scope of those data.

That creates a comparison between Tesla’s operating data and a broad national estimate rather than between equivalent vehicles driven on the same roads, at the same times and in the same weather and traffic conditions. Vehicle age, road type, geography, driver population and where the assistance system is engaged can all affect the exposure represented by a mile.

The manual-Tesla comparison narrows one source of variation because both groups use vehicles from the same manufacturer, but it still does not establish that the driving contexts are matched. Drivers may engage assistance more often on some roads or in some conditions than others. The reported intervals do not provide the contextual breakdown needed to isolate the system’s contribution from those usage patterns.

Collision definitions also shape the result. Airbag deployment is a concrete dividing line between Tesla’s major and minor categories, but it is not a complete measure of injury severity or crash risk. The five-second attribution window is comparatively inclusive because it can count a collision after the system has ceased to be active, while counting incidents regardless of responsibility avoids removing crashes attributed to another road user.

Real-world scale and experimental control answer different questions

Fleet data provide extensive real-world exposure, including traffic interactions that are difficult to reproduce in a laboratory. Their limitation is control: sensing performance, control decisions, driver supervision and the surrounding traffic environment vary together. A controlled study can isolate variables more effectively, but it usually covers fewer miles and a narrower set of conditions.

Research into partial vehicle automation has consequently produced different findings across methods. Controlled and naturalistic studies can examine attention, system use and driver behavior, while crash databases capture real outcomes but often cannot isolate causation. A 2026 cross-methodological review found that real-world data tend to indicate net safety benefits from partial automation, while also identifying confounding factors and uncertainty over whether the benefit comes from automated steering and speed control or associated safety systems.

Tesla’s latest update therefore supplies a meaningful internal trend: its reported major-collision interval rose by roughly 700,000 miles from the previous report. Establishing how much of that change comes from FSD Supervised itself would require matched exposure data that account for where, when and under what conditions assisted and manual miles were driven.

By Jonathan Barrett — Editor for AMI’s future mobility and autonomous systems section, with two decades covering robotics, e-mobility, drone-vehicle convergence, and transport mechanical systems.