Tesla rolls out FSD v14.3 updates

Tesla’s latest Full Self-Driving update is less about one flashy trick and more about tightening the whole driving stack. FSD v14.3 started rolling out in April 2026, with follow-up builds v14.3.1 and v14.3.2 landing after that for U.S. vehicles with Hardware 4. The big idea is simple — make the car react faster, see weird situations better, and train on the rare edge cases that usually break driver-assist systems. Tesla is also clearly treating this as more than a consumer feature update, because the same work now feeds its Robotaxi software path. ### What actually shipped? The release notes point to three concrete changes. First, Tesla says it upgraded the reinforcement-learning stage that trains the FSD neural network. Second, it upgraded the vision encoder so the car does better in rare and low-visibility situations and gets stronger 3D geometry understanding. Third, it rewrote the AI compiler and runtime with MLIR — a machine-learning compiler framework — which Tesla says improves model iteration speed and delivers 20% faster reaction time. (assets-ir.tesla.com) ### Why does the 20% reaction number matter? Because most of the hard problems in assisted driving are not normal lane-following. They’re the ugly, late-breaking moments — a car blowing through right of way, a weird object leaning into the lane, a compound traffic light on a curved road, or a temporary sensor degradation. In those moments, shaving processing delay matters more than adding another marketing label. (notateslaapp.com) Tesla is basically saying v14.3 is faster not just in the abstract, but in the exact category of scenarios where hesitation turns into a disengagement or a near miss. ### What kinds of edge cases got attention? The release notes are unusually explicit here. Tesla calls out better response to emergency vehicles, school buses, right-of-way violators, and other rare vehicles. It also says the system improved traffic-light handling at complex intersections with compound lights, curved roads, and yellow-light stopping. On top of that, it added better handling for unusual objects extending, hanging, or leaning into the path, plus improved behavior around small animals. (notateslaapp.com) That is a very edge-case-heavy list — which tells you where Tesla thinks the remaining pain points are. ### Where is this rolling out? Right now, this is a U.S. rollout for Models S, 3, X, Y, Cybertruck, and it is limited to Hardware 4 on the published v14.3.1 and v14.3.2 notes. Tesla’s support pages also framed v14 as a supervised feature set for owners in the U.S., Puerto Rico, Mexico, and Canada during the trial window, but the specific newer v14.3.x builds shown in release notes are U.S. (notateslaapp.com) HW4 builds. That hardware split matters because HW3 cars are still on a different path. ### Why bring Robotaxi into this? Because Tesla did, directly. In v14.3.2, the company added a new note saying it unified the model between Actually Smart Summon, FSD, and Robotaxi for more capable and reliable behavior. That is the clearest signal in the rollout. Tesla is no longer presenting consumer FSD and robotaxi software as mostly separate stories. It is presenting them as one shared model with different operating modes. (notateslaapp.com) ### Does this make the car autonomous? No — and Tesla still says that plainly. Its support pages keep calling the product “Full Self-Driving (Supervised)” and say active driver supervision is required. The car can handle a lot, but Tesla is still shipping this as advanced driver assistance, not a hands-off consumer autonomy launch. That distinction matters even if the company is using these updates to harden the software for robotaxi operations behind the scenes. (notateslaapp.com) ### What’s the bottom line? v14.3 looks like Tesla’s “make the stack sharper” release. Not a miracle jump, but a serious one — faster reactions, better handling of weird scenes, and a tighter link between owner cars and robotaxi software. The catch is that it is still supervised and still hardware-gated. But if you want to know how Tesla thinks autonomy gets built, this update is the answer: grind through the rare failures, speed up the model, and merge everything into one system. (tesla.com)