Tesla is getting ready to roll out FSD V14. With FSD V13.2.6 still expanding to AI4 cars, the company has been making steady improvements since launching FSD V13 just over two months ago.
FSD V13 introduced key features like starting FSD from Park and Reverse, plus parking at a destination.
It also brought full-resolution video input at 36Hz using AI4 cameras and improved decision-making with the new Cortex supercomputer.
So, what’s next? Tesla has shared a preview of what’s coming in FSD V14.
Auto-Regressive Transformers
The main update in FSD V14 is the use of auto-regressive transformers. This sounds technical, but here’s what it means.
Auto-Regressive
This method processes sequences over time and uses past data to predict what comes next. Think about finishing a sentence, you use previous words to guess the next ones.
For FSD, this means analyzing a sequence of camera images to track pedestrians and predict their movement based on what’s happening around them.
Right now, FSD reacts to what it sees. Soon, it will anticipate what’s coming, much like a human driver.
Transformers
Transformers help the system decide what’s most important in a time sequence. They highlight key details for better predictions.
For example, a blinking turn signal is more useful than a car’s color when figuring out a lane change.
Why This Matters
With auto-regressive transformers, FSD will get better at predicting how people and cars move. This will help it make smarter, more human-like decisions.
Larger Model and Context Size
Tesla’s VP of AI, Ashok Elluswamy, said FSD V14 will have a larger model and bigger context size. Context is the history of what the car remembers to make decisions.
The AI4 computer’s memory limits how much context FSD can store, but Tesla is still making improvements.
The upcoming update should expand model and context size by 3x.
Audio Input
Tesla has been collecting audio data in past FSD versions. Now, FSD V14 will start using that data.
The first use will be for detecting emergency vehicles, but later, it could respond to other sounds like car crashes, loud noises, or honking.
This makes driving more like how humans react to their surroundings.
When Will FSD V14 Arrive?
Tesla hasn’t given a release date. Ashok previously said that audio input would be used in FSD V13.4, but during the latest earnings call, Tesla mentioned it as a feature for V14. This suggests Tesla might skip V13.4 entirely.
With the Tesla Robotaxi network launching in Texas this June, FSD V14 could be the version powering those autonomous taxis.
We’ve launched an official Tesla Owners Online Forum. Join early and claim your "Founding Fuel" badge.
