Tesla Launches FSD v14 Lite in South Korea
Tesla has officially launched the FSD (Supervised) v14 Lite in South Korea.
Vehicles in the U.S. Model 3/Y (HW3/AI3) that have activated FSD are beginning to receive updates. This version adapts the driving behavior of HW4 v14 to the HW3 hardware through knowledge distillation.
In the market mechanism, Tesla prioritizes the push to HW3 vehicles in South Korea, quickly approved under the Korea-U.S. FTA provisions. Vehicles from China still require local regulatory approval. Early feedback will drive subsequent expansion, allowing older hardware users to continue benefiting from the FSD subscription value.
Source: Public Information
ABAB AI Insight
Tesla previously promised that HW3 v14 Lite would roll out in the U.S. before expanding internationally. Ashok Elluswamy confirmed that this version addresses the long-standing stagnation of HW3 since v12.6.4 in early 2025 through reinforcement learning (RL) and offline model distillation of HW4 v14 capabilities. HW3 users have been dissatisfied with subscription performance gaps for over a year.
On the capital path, Tesla retains millions of HW3 vehicles (approximately 4 million globally) as paying users through a tiered subscription strategy (the Lite version may be lower priced), avoiding hardware upgrade costs. Meanwhile, testing in markets like South Korea accelerates global regulatory approvals, paving the way for large-scale Robotaxi deployment, shifting resources from hardware iteration to software distillation and data feedback loops.
Similar to Elon Musk's early Autopilot promises, HW3 was marketed in 2019 as "future-proof." The Lite version now resembles performance optimizations in mobile chips (like Apple's A series distillation). Tesla is in a transitional phase from supervised FSD to unsupervised, with HW3 gaining core functionalities like parking and speed curves, albeit with limited responsiveness.
Essentially, this represents a reconstruction of the industry chain under technological substitution: computational hardware bottlenecks are addressed through AI distillation, allowing old hardware to approach new functionalities without replacement. The mechanism relies on vast amounts of real driving data enabling neural networks to transfer advanced behaviors on low computing power, lowering user switching barriers while concentrating capital on data training and regulatory lobbying.
ABAB News · Cognitive Laws
- When hardware is locked, distillation is the true lifeline.
- Retaining existing users outweighs pleasing new hardware.
- Commitments to iteration lag behind, while data feedback loops advance.