Tesla reaches 10B FSD supervised miles

Tesla just put a huge number on the board — 10 billion miles driven with Full Self-Driving (Supervised). That matters because Elon Musk spent January saying “roughly” that much data was needed for safe unsupervised driving. Now the counter has reached the mark. But the gap is still the same one Tesla has wrestled with for years: a giant pile of supervised miles is not the same thing as proving a car can drive alone safely in the wild. (tesla.com) ### What happened today? Tesla’s FSD safety page now shows 10,024,620,946 miles driven, including 3,766,563,904 city miles. The company also highlighted the milestone on social media on May 4. This is a cumulative fleet number — not one software version, not one geography, and not an unsupervised robotaxi total. (tesla.com) ### Why is 10 billion such a loaded number? Because M(tesla.com)eeded roughly 10 billion miles of training data for “safe unsupervised self-driving,” arguing that real roads have a “super long tail of complexity.” So hitting the number is not just a nice round stat — it lands directly on Tesla’s own public benchmark. (forbes.com)for-safe-self-driving/)) ### What exactly are these miles? They are supervised miles. Tesla’s own FSD pages say the system requires active driver supervision and does not make the vehicle autonomous. Tesla sells it as a $99-per-month feature that can handle navigation, lane changes, turns, parking, and similar tasks, but the human is still legally and practically the fallback driver. That distinction is the whole story here. (tesla.com) ### So does this mean Tesla can flip to unsupervised now? Not automatically. A mileage threshold is a training and confidence argument, basically a claim that the fleet has seen enough weird edge cases. But safety validation is not just about how much data you have. It is also about where the miles happened, how often humans intervened, how the system behaves in rare failures, and whether regulators accept the evidence. (tesla.com)— but data volume alone does not settle the question. (tesla.com) ### Why do critics keep pushing back? Because supervised miles are a messy metric. If a human is watching, taking over, and correcting mistakes, the number mixes software performance with human backup. That is useful for training, but it is not the same as a clean, independent test of driverless safety. Think of it like counting hours with a student pilot and treating them as equivalent to solo flight — related, but not interchangeable. (tesla.com) ### What are regulators looking at right now? NHTSA is still digging into Tesla’s FSD in reduced-visibility conditions. In March 2026, the agency opened an Engineering Analysis covering about 3.2 million vehicles equipped with FSD, focused on whether the system properly detects degraded visibility and warns the driver in time. The case summary lists nine crashes, including injury incidents, tied to those concerns. That (tesla.com)ut it does show why the milestone does not end the argument. (static.nhtsa.gov) ### How does this fit Tesla’s broader 2026 push? Tesla’s Q1 2026 update framed FSD progress as part of a bigger autonomy buildout. The company said it received approval for FSD (Supervised) in the Netherlands in April and launched unsupervised Robotaxi rides in Dallas and Houston the same month. So the 10 billion figure arrives as Tesla tries to connect three things into one sto(static.nhtsa.gov)peration. (assets-ir.tesla.com) ### What should readers take from the number? It is real progress, and it is also a marketing number. Tesla has clearly built extraordinary scale in real-world driving data. But the hard part was never just collecting miles. The hard part is showing that those miles translate into robust behavior in the weird, low-frequency failures that matter most — glare, fog, bad markings, strange human behavior, the stuff that breaks neat demos. (tesla.com) ### Bottom line? Tesla hit the benchmark it chose. That is important. But the milestone does not answer the only question that matters — whether supervised fleet learning is enough to earn trust for truly unsupervised driving. (tesla.com)