Tesla driving score: what is left after Tesla closed Safety Score to third-party apps?

What is Tesla's Safety Score, and who can see it?

Tesla's Safety Score turns driving data measured by the car into a number from 0 to 100, where higher means safer. According to Teslarati's April 2025 report, the current beta, version 2.2, looks at seven factors: hard braking, aggressive turning, unsafe following, excessive speeding, late-night driving, forced Autopilot disengagement and unbuckled driving. The same report says Tesla built the formula from more than 22 billion miles of fleet data and dropped forward collision warnings from it in 2.2.

The score exists mainly for Tesla's own insurance. Teslarati reports that Tesla Insurance uses it to set premiums in the US, except in California, where privacy law prohibits using real-time driving data for premiums. Tessie's help page adds that enrolment is limited to Tesla Insurance and FSD Beta customers. In other words, it is a score Tesla built for its own pricing and versions as it sees fit.

When did third-party access end, and what does that mean today?

Tessie's help article on Tesla's Safety Score carries a one-line note: as of December 2023, Tesla no longer supports the score in third-party apps. Apps like Tessie used to show it; they no longer can. The page gives no reason, and we won't guess one.

The practical upshot: if an app shows you a number it calls your Tesla score today, it is either copied by hand from the Tesla app or a different number the app computes itself. The second is legitimate, as long as the number has its own name and its formula is published.

What does Fleet Telemetry expose?

According to Tesla's developer documentation, Fleet Telemetry lets the car stream data directly to a server the developer runs, with data sent as often as every 500 milliseconds. The driver has to pair the service's virtual key with the car in the Tesla app, and the car needs firmware 2024.26 or newer (some older cars have different minimums).

One detail matters a lot: Tesla's field list has no ready-made hard-braking event, harsh-acceleration event or driving-score field, and no field for the road's posted speed limit. Fields such as ForwardCollisionWarning report the sensitivity the driver chose in the settings, not an event. Events have to be derived from the acceleration stream, and what makes a score trustworthy is that this derivation is written down.

The fields that matter for a behaviour index, in Tesla's own descriptions:

• LongitudinalAcceleration: acceleration along the car, m/s² (braking and accelerating)
• LateralAcceleration: sideways acceleration, m/s² (cornering)
• VehicleSpeed: vehicle speed; Tesla documents the unit as miles per hour
• Gear: the gear reported by the drive inverter
• BrakePedalPos and PedalPosition: master-cylinder pressure and accelerator pedal position
• Location: latitude and longitude. The index does not need it; below we explain why it never goes to an insurer

How a transparent behaviour index is computed: EVVisor's method

EVVisor's Driving Index is a number from 0 to 100, where higher means calmer driving. The published model version is fdi-1.0.0. Each factor's raw value is scaled to 0–100, multiplied by its weight and summed. The index is never shown without its factors. A reading only counts if it stays past the threshold for at least 300 milliseconds while the car is doing at least 5 km/h.

To be upfront about speeding: because Fleet Telemetry does not provide the road's speed limit, fdi-1.0.0 does not measure exceeding the legal limit. It measures time spent above 120 km/h. That is a proxy, and it misses urban speeding. When road matching is added, it will be a new model version; old scores will not change silently.

With too little data, the index is flagged provisional: under 50 km in the period, or fewer than 2 samples per km, and a 1–2 second braking event cannot be seen reliably. Thresholds will be tuned on pilot data, and every change gets a new version number so an insurer can reproduce a past score with the same model. The factors and weights in fdi-1.0.0:

• Hard braking, weight 25%: longitudinal acceleration below −2.94 m/s² (0.3 g); events per 100 km
• Aggressive turning, 15%: lateral acceleration above 3.92 m/s² (0.4 g) either way; events per 100 km
• Speeding, 25%: share of moving time above 120 km/h (proxy)
• Late-night driving, 15%: share of km driven 00:00–04:00 Istanbul time
• Excessive acceleration, 10%: longitudinal acceleration above +2.94 m/s² (0.3 g); events per 100 km
• Trip context, 10%: share of km driven on trips shorter than 5 km

Why this is not Tesla's Safety Score, and why the numbers won't match

The Driving Index is not a copy or an estimate of Tesla's score, and it is never presented as one. A driver who puts the two numbers side by side will see different results. That is not a bug; it is what two different formulas produce:

• Different factors: Tesla's version 2.2 also looks at following distance, forced Autopilot disengagement and seatbelt use (Teslarati); fdi-1.0.0 does not, and adds trip context.
• Different night window: according to Electrek, Tesla's version 2.0 weighted late-night driving by hour between 10 pm and 4 am; we count only 00:00–04:00.
• Different speeding rule: according to Electrek, version 2.0 penalised driving above 85 mph (about 137 km/h); our proxy threshold is 120 km/h.
• Different weights and scaling, and Tesla sets its own version numbers when it changes its formula.
• Different name: ours is the Driving Index, and it is never displayed as another brand's score.

Consent, KVKK and location: who keeps the data?

The driver shares the data. Under Article 5 of Turkey's Personal Data Protection Law (KVKK, No. 6698), personal data may not be processed without the person's explicit consent, apart from the exceptions the law lists. The law defines explicit consent as consent on a specific subject, based on information and given freely. Article 11 gives the right to learn whether data is processed and, under the law's conditions, to ask for its deletion.

When the driver withdraws, streaming stops at the car: Tesla's documentation says that if a required authorisation scope is revoked, the telemetry configuration is removed from the vehicle. At EVVisor, what reaches an insurer is the score, its factors, event counts and data quality; raw location never goes to an insurer. One driver's index is shared with an insurer only with that driver's explicit consent for that specific sharing.

Status: in pilot, no public scores

The Driving Index is in pilot. Streaming data from the pilot cars starts in early October 2026; until then, any sample score you see is built on sample data. We do not publish public scores, rankings or averages; when we do, they will come with the sample size and model version.

The index does not set a price and does not promise a discount. Whether an insurer uses this kind of data, and how it affects a premium, is the insurer's decision.

Frequently asked questions

Can I see Tesla's Safety Score in a third-party app?

No. According to Tessie's help centre, Tesla has not supported the score in third-party apps since December 2023. It is visible in the Tesla app, for drivers enrolled in the programme.

Will EVVisor's Driving Index match Tesla's number?

No. The factors, thresholds, night window and weights differ. The index does not try to predict Tesla's score; it is computed with its own published formula and carries its model version.

Does Fleet Telemetry report hard-braking events?

No. Tesla provides longitudinal and lateral acceleration in m/s²; braking, cornering and acceleration events have to be derived from that stream with thresholds. In fdi-1.0.0 the hard-braking threshold is 0.3 g and the cornering threshold 0.4 g.

Does my location go to the insurer?

No. An insurer receives only the score, its factors, event counts and data quality, and one driver's index only with that driver's explicit consent. Raw location stays with the driver.

Will a good index make my insurance cheaper?

We can't say that. The index describes driving; it does not set a price. The insurer decides whether to use the data and how it affects a premium.