Tesla's Elluswamy: FSD v15 to become a 10-billion-parameter neural net
Tesla AI chief Ashok Elluswamy says the next-gen FSD v15 will scale its neural network to 10 billion parameters — 10x today's size — to sharply improve hazard prediction, reaction time, and collision avoidance. Robotaxi vehicles are already running early v15 builds in live operating zones, making this the next big test of Tesla's self-driving reliability.
Elluswamy described v15 as more than an incremental update — an architectural leap that ties prediction and control more tightly together. The stated goal, in his words: "even earlier prediction of hazards, even faster reaction time and overall significantly better safety and collision avoidance."
The centerpiece is a 10x jump in model size, to 10 billion parameters. Tesla says this scale enables a form of 'active reasoning' that helps the car anticipate hidden hazards and handle chaotic scenarios such as multi-lane roundabouts.
v15 is already in real-world validation: Robotaxi vehicles are running early builds across active operating zones, gathering data ahead of a planned consumer rollout in late 2026 or early 2027.
Tesla previously added Automatic Collision Evasion — which lets FSD briefly take over steering to dodge an imminent crash — in FSD 14.3.9 on September 4. Unlike that discrete safety feature, v15 represents a more fundamental overhaul of the network's own prediction and decision-making capacity.
Summaries are prepared by the Tesla Briefing editorial team and may not capture every nuance of the original reporting. You are solely responsible for your own investment decisions.