Tesla's latest Full Self-Driving (FSD) software, version 14.3, is rolling out with significantly enhanced reasoning capabilities, aiming to provide "superhuman judgement" in complex driving scenarios. This update builds on existing "superhuman perception and reflexes," according to a recent social media post by user "jimmah." The new version also boasts a 20% faster AI reaction time, a critical improvement for autonomous driving systems due to an MLIR-based AI compiler rewrite.
Tesla AI Chief Ashok Elluswamy confirmed that reasoning features, initially expected in v14.3, have already begun shipping in the current v14.2 build for specific high-complexity tasks like construction navigation and parking. This expansion moves beyond mere destination handling to encompass all driving behaviors, addressing edge cases previously challenging for the system. CEO Elon Musk has described v14.3's target behavior as "sentient," implying the car will think and reason more like a human driver rather than simply following programmed rules.
Beyond reasoning, FSD v14.3 introduces a new pothole avoidance feature, designed to protect vehicle components by actively steering around road hazards. The update also includes a more sensitive driver monitoring system with improved eye gaze tracking and better handling of eyewear, aiming to reduce false alerts and enhance user experience. These enhancements are part of a broader effort to make the FSD experience more competent and less prone to disengagements.
The v14.3 update is primarily targeting vehicles equipped with Hardware 4 (HW4), with a "FSD v14 Lite" version for Hardware 3 (HW3) owners anticipated in mid-2026. This phased rollout allows for targeted deployment and data collection across the fleet. While many users report smoother and more confident maneuvers, some community discussions indicate ongoing challenges, such as early turn signal activation or occasional unexpected detours, highlighting the complexity of achieving full autonomy.
The integration of expanded reasoning is considered a crucial step towards Tesla's long-term goal of fully unsupervised Full Self-Driving. By enabling the system to apply deliberate, multi-step thinking to unforeseen situations, Tesla aims to close the gap between current Level 2 assistance and a truly autonomous driving experience. This continuous development underscores the company's iterative approach to advancing its AI capabilities in real-world conditions.
