In a highly anticipated move, Tesla has officially initiated the rollout of its Full Self-Driving (FSD) beta version 12 beyond its employee base. Initially released to a limited number of employees last month, the FSD beta v12.1 is now being extended to an initial wave of cars, encompassing over 15,000 personally-owned Tesla employee vehicles, according to software update tracker Teslascope.
This latest version of the FSD beta has been eagerly awaited throughout the year, promising significant improvements over its predecessors by shifting to neural network-based operations designed for city driving. Speculation is now rife about the possibility of the software being deployed to the wider public, with some industry observers predicting a potential release as early as mid-January.
Teslascope recently shared a screenshot of release notes for FSD beta v12.1 from Tesla employees. The notes, currently brief, are expected to undergo additions before any public release to non-employees. Notably, those participating in the initial wave of the rollout, numbering between 16,500 and 20,000 vehicles, are bound by strict non-disclosure agreements, with violations risking termination or removal from the FSD beta program.
FSD beta v12 represents a significant shift, transitioning the autosteer on city streets system to a single stack end-to-end neural network trained on millions of video clips. This groundbreaking approach eliminates the need for over 300,000 lines of human-written C++ code.
Tesla CEO Elon Musk showcased an early build of FSD v12 in a live stream in August, and last month, he confirmed the version’s deployment to some employees. Musk has also indicated the removal of the “beta” distinction from FSD v12, though its current status remains unclear, with the Teslascope screenshot still displaying the term.
Earlier this month, Tesla’s artificial intelligence (AI) account on X extended a recruitment call, encouraging individuals to apply to contribute to the FSD beta. The company emphasized its pursuit of “next-generation autonomy” built on a single foundation video network that drives the car, utilizing data from one of the world’s largest training clusters.