Tesla FSD gets praised

Tesla’s driving software is still supervised, which means the human in the seat is legally and practically the backup driver, but a new April 2026 update is getting unusually strong praise from early testers for feeling quicker and less awkward in everyday traffic. Tesla began rolling out Full Self-Driving (Supervised) v14.3 on software build 2026.2.9.6 to Hardware 4 vehicles on April 7. (electrek.co) The biggest number in the update is Tesla’s claim of “20% faster reaction time,” and Tesla says that gain came from rewriting the artificial intelligence compiler and runtime with Multi-Level Intermediate Representation, a tool for turning neural-network code into something the car’s computer can execute faster. Tesla tied that rewrite directly to quicker responses and faster model iteration in its v14.3 release notes. (electrek.co) Tesla also says it upgraded the training stage called reinforcement learning, which is the part where the system gets rewarded for better choices the way a student improves with repeated feedback. In the same notes, Tesla says that change improved “a wide variety of driving scenarios” rather than one narrow test case. (evshift.com) Another change is the vision encoder, which is the software that turns raw camera video into the car’s first draft of what the road looks like. Tesla says the new encoder is better in rare events, low-visibility scenes, three-dimensional geometry, and traffic-sign understanding. (evshift.com) That helps explain why early owner videos are focusing on night driving and sudden hazards instead of just smooth highway clips. Tesla’s own notes say v14.3 improves handling of rare objects leaning into the lane, better response to emergency vehicles and school buses, and better behavior around small animals. (electrek.co) The owner praise is not just about speed. Tesla specifically says it reduced unnecessary lane biasing and minor tailgating, which are the little habits that make automated driving feel twitchy or rude even when it does not fully fail. (evshift.com) Parking got special attention too, and that matters because low-speed maneuvers are where robotic driving often looks most hesitant. Tesla says v14.3 increased the decisiveness of parking-spot selection and maneuvering, and one April 7 first-impressions review said four parking attempts were “flawless” and noticeably quicker than v14.2.2.5. (electrek.co) (teslarati.com) The same first-impressions review said yellow-light behavior improved because the car was less likely to stab the brakes when continuing through the intersection was the safer move. That lines up with Tesla’s release notes, which say v14.3 improved traffic-light handling at complex intersections, curved roads, and yellow-light stops. (teslarati.com) (evshift.com) The praise comes with a clear asterisk: the hardest problem is still not just seeing the road but choosing the right route every time. The same April 7 review said navigation and routing remain Tesla’s biggest autonomy challenge, especially when the car’s path conflicts with the obvious simplest way to a destination. (teslarati.com) Tesla is also already pointing past this version. In the v14.3 notes, Tesla listed “expand reasoning to all behaviors beyond destination handling” and “add pothole avoidance” as upcoming improvements, which suggests the company still sees the current build as a step in a longer rewrite rather than a finished product. (evshift.com) So the story here is not that Tesla suddenly solved self-driving in one patch on April 7, 2026. The story is that v14.3 appears to have made the car feel more natural in the small moments drivers notice first — braking, lane choice, parking, and low-light surprises — while the bigger promise of beating human safety is still a claim waiting for broad real-world proof. (electrek.co) (teslarati.com)