Tesla sets NYC–LA Cannonball record

Tesla’s latest Full Self-Driving milestone is a coast-to-coast brag with real technical weight — but also a big asterisk. Dan Burkland says his Tesla Model 3 drove from Manhattan’s Red Ball Garage to the Portofino Hotel in Redondo Beach in 49 hours, 55 minutes, and 57 seconds with zero disengagements. If that holds up, it resets the benchmark for a supervised Tesla FSD Cannonball run. But this is still a self-reported claim, and the comparison everyone wants to make is messier than it looks. ### What actually happened? Burkland posted that his team completed the classic New York-to-Los Angeles Cannonball route in a refreshed Model 3 running FSD v14.3.2, with the software handling the full trip and even the parking at Supercharger stops. That last detail matters more than it sounds — parking lots have been one of FSD’s softer spots, so folding charging-stop maneuvering into the “zero disengagements” claim raises the bar beyond just highway cruising. (evxl.co) ### Why is everyone comparing it to Alex Roy? Because Alex Roy’s January run was the previous headline benchmark. Roy and a team of autonomy testers took a 2024 Model S from Los Angeles to New York in 58 hours and 22 minutes with zero interventions, using HW4 hardware and FSD v14.2.2.3. That run got attention partly because it finally matched, in broad form, Elon Musk’s old promise that a Tesla would one day make the trip end to end on its own. (evxl.co) ### So is this really an eight-hour improvement? Sort of — but not cleanly. Burkland’s claimed time is 8 hours and 27 minutes quicker than Roy’s. The catch is that the traditional New York-to-LA Cannonball route is roughly 2,800 to 2,900 miles, while Roy’s January trip covered 3,081 miles. That distance gap alone can explain a meaningful chunk of the time difference before you even get to software, weather, traffic, or charging strategy. (thedrive.com) ### Why does the direction matter so much? Because westbound in May is not the same test as eastbound in January. Roy’s team dealt with winter conditions, including snow, ice, and slush, plus detours. Burkland’s run, at least from what’s been posted so far, did not report precipitation. So this is not one stopwatch proving Tesla suddenly got dramatically better in every condition. It’s a faster run under friendlier conditions on a shorter route. (evxl.co) ### What changed in the software? The big shift is v14.3. Tesla’s release notes for that branch say the company rewrote the AI compiler and runtime around MLIR and claimed a 20% faster reaction time. Tesla also said v14.3 improved low-visibility handling, rare-scenario understanding, and parking behavior. Burkland’s car was on v14.3.2, two revisions beyond that base release, so the run fits a broader story: Tesla is still pushing meaningful point-release changes, not just cosmetic tweaks. (evxl.co) ### Does this mean Tesla has solved self-driving? No — because FSD is still supervised. A zero-disengagement road trip is impressive, but it is not the same thing as a car being legally or operationally autonomous in all conditions. Coast-to-coast highway driving is a very specific test. It says a lot about endurance and consistency. It says less about edge cases in cities, bad weather, weird construction zones, or failure handling when visibility falls apart. (electrek.co) ### Why are regulators still on Tesla’s case? Because the federal safety picture is moving the opposite way from the hype cycle. In March, NHTSA escalated its probe into Tesla’s FSD system to an engineering analysis — its highest level of defect investigation — focused on reduced-visibility conditions. That investigation covers about 3.2 million vehicles, and regulators have said reviewed crashes showed the system sometimes failed to detect impaired camera visibility or warn drivers early enough. (thedrive.com) ### What’s the real takeaway? Basically, this run matters as a capability signal, not as final proof. Tesla FSD appears to be getting better fast enough that enthusiasts can now plausibly chase coast-to-coast “zero intervention” records. But the system is still being judged in two worlds at once — one world of viral demos and benchmark drives, and another world of formal safety scrutiny. Both are real. (evxl.co) (techcrunch.com)