According to Sawyer Merritt on X, Tesla has released FSD v14.3 with AI-centric upgrades including a ground-up rewrite of the AI compiler and runtime using MLIR that delivers roughly 20% faster reaction times and accelerates model iteration, alongside improvements to the neural network vision encoder and an upgraded reinforcement learning stage trained on hard fleet-sourced examples (as reported by Sawyer Merritt). According to Sawyer Merritt, v14.3 also enhances handling of emergency vehicles, school buses, complex traffic lights, rare objects intruding into the path, and reduces unnecessary disengagements by maintaining control during temporary system degradations (as reported by Sawyer Merritt). According to Sawyer Merritt, Tesla’s next updates will expand reasoning to all behaviors beyond destination handling, add pothole avoidance, and improve the in-cabin driver monitoring system with better eye gaze tracking, eyewear handling, and higher accuracy in variable lighting—signaling deeper end-to-end autonomy capabilities and safety-focused computer vision enhancements (as reported by Sawyer Merritt).