Tesla Sets 10 Billion Mile Goal for Unsupervised FSD; Ashok Confirms Reasoning is Here

Elon Musk posted on X with an update about the amount of data Tesla believes is necessary to reach superhuman safety for Unsupervised FSD. Tesla’s Director of AI, Ashok Elluswamy, also confirmed that some reasoning capabilities—often described as the holy grail of AI decision-making—have already been delivered in FSD, with additional reasoning features planned.

10 Billion Miles: 7 and Counting

In his Master Plan Part Deux, Elon originally cited a figure of 6 billion miles as the dataset size required to secure worldwide regulatory approval for true autonomy via Unsupervised FSD. That target implied a system that is exponentially safer than human drivers and continues improving through the march of 9’s to address edge cases and real-world complexity.

That figure has now been updated.

Roughly 10 billion miles of training data is needed to achieve safe unsupervised self-driving. Reality has a super long tail of complexity.

— Elon Musk (@elonmusk) January 8, 2026

The total Elon and Tesla are targeting refers to real-world miles. While Tesla and other autonomy developers use simulated data for training, real-world driving data is essential to handle rare edge cases that occur in the real world.

Where is Tesla Now?

According to Tesla’s FSD Safety Hub, the fleet has accumulated approximately 7.1 billion miles on FSD. That leaves roughly 2.9 billion miles to reach the newly stated 10 billion-mile target.

With FSD adoption slowly increasing and the fleet expanding annually, that remaining distance may be covered relatively soon. In March 2025 Tesla reported 3.6 billion miles on FSD; the fleet now sits at about 7.1 billion miles. At the current pace, Tesla could reach 10 billion miles within the next six months.

This candid adjustment from Elon underscores how difficult it is to chase the long tail of edge cases: getting to 99% is straightforward, but progressing to 99.9%, 99.99%, and beyond becomes exponentially harder. This observation followed a similar note of caution from Elon and Ashok directed at NVIDIA after the debut of Alpamayo at CES 2026.

The near-infinite variety of edge cases—from unmapped routes and sudden changes, to erratic human or animal behavior and challenging weather—remains a central challenge.

Reasoning with V14

To close the gap between 7 billion and 10 billion miles, Tesla is not solely relying on more data; it is changing how FSD reasons and makes decisions.

Ashok stated that FSD V14.2, which is currently rolling out to customer vehicles, already contains some early implementations of reasoning capabilities.

Some elements of reasoning such as navigation route changes during construction, parking options have already shipped in 14.2. More and more reasoning will ship in Q1.

— Ashok Elluswamy (@aelluswamy) January 9, 2026

What is Reasoning Really?

Until now, FSD has mostly acted as a fast reflex system—spotting a red light or an obstacle and reacting. Reasoning enables the vehicle to simulate multiple potential futures before acting, planning ahead about route choices and maneuvers in real time much like a human does.

Put simply, reasoning improves human-like driving: rather than halting where a road is abruptly closed, the car can determine it needs to divert, observe the surroundings, and select an alternate path.

In the example Ashok used about parking, reasoning means the vehicle can choose a safe spot near a curb, avoid stopping at a driveway, and not simply pull over at the final pinned location.

The Takeaways

Elon’s 10-billion-mile estimate recognizes how challenging real-world driving complexity is. Ashok’s update indicates that a practical path forward—Vision-Language-Action models that think rather than only react—is already underway rather than being purely theoretical.