He Xiaopeng Praises Tesla FSD 14.2: What XPeng's CEO Sees in Tesla's Vision-Only Stack

When XPeng CEO He Xiaopeng publicly praised Tesla's Full Self-Driving (FSD) software, it sent ripples through the autonomous-driving industry. His comments specifically highlighted the progress Tesla has made with its vision-only neural-network approach — a stack that Tesla has iterated on through successive releases. For EV buyers and tech watchers alike, the endorsement from a direct competitor is significant: it signals that Tesla's camera-based architecture has reached a maturity level that even rival engineering teams acknowledge.

But praise from an industry peer doesn't mean the system is flawless — or that it's "self-driving." Here's a precise, spec-driven breakdown of what Tesla FSD (Supervised) actually does in its latest iterations, what it costs, where it falls short, and how it stacks up against XPeng's own XNGP system and other Level 2 competitors.

What Tesla FSD (Supervised) Actually Does

Tesla's Full Self-Driving capability — officially branded FSD (Supervised) — is an advanced driver-assistance system that extends beyond basic Autopilot. Where standard Autopilot handles lane centering and adaptive cruise control on highways, FSD (Supervised) adds:

• Navigate on Autopilot: Automated highway on-ramp to off-ramp routing, including lane changes and interchange navigation
• Auto Lane Change: Initiates lane changes based on navigation route and surrounding traffic
• Autosteer on City Streets: Traffic light and stop sign recognition, intersection navigation, and turns on surface streets
• Smart Summon: Vehicle navigates parking lots to the driver's location (within geofenced constraints)
• AutoPark: Parallel and perpendicular parking automation

Under the hood, Tesla's system relies entirely on a camera-only (Tesla Vision) sensor suite. Current Tesla vehicles (Model 3, Model Y, Model S, Model X, and Cybertruck) ship with a camera array — typically 8 external cameras providing 360-degree coverage — paired with Tesla's custom HW4 (Hardware 4) AI inference computer. Tesla removed ultrasonic sensors and radar from new production vehicles starting in 2022–2023, betting fully on neural-network-driven perception from camera data.

The SAE Level 2 Reality: Not Self-Driving

Tesla's own owner's manual states plainly: the driver is responsible for the vehicle at all times. The interior cabin camera monitors driver attention, and the system will issue escalating warnings — visual, audible, and eventually a forced pull-over — if it detects inattention. Ignoring these warnings results in temporary or permanent suspension of Autosteer access.

How Much Does Tesla FSD Cost in 2026?

Tesla has adjusted FSD pricing several times. As of current 2026 offerings, buyers have two paths:

Basic Autopilot (Traffic-Aware Cruise Control + Autosteer on highways) is included standard on all new Tesla vehicles at no additional cost. Enhanced Autopilot — which adds Navigate on Autopilot, Auto Lane Change, AutoPark, and Smart Summon — is available as a $6,000 one-time purchase. FSD (Supervised) builds on Enhanced Autopilot by adding city-street driving capability.

Important: FSD (Supervised) is not transferable between vehicles if you buy a new Tesla. Tesla has occasionally offered promotional FSD transfers during end-of-quarter sales pushes, but this is not standard policy. The subscription model is often the smarter financial choice unless you plan to keep the vehicle for 7+ years.

Which Tesla Models Include FSD Capability?

FSD (Supervised) is available as a purchase or subscription on every current-production Tesla model, provided the vehicle has Hardware 3 (HW3) or newer. All vehicles built since approximately April 2019 include HW3 or HW4. Older vehicles with HW2.5 or earlier cannot run the latest FSD stack and would require a hardware retrofit (which Tesla has offered in limited cases).

Real-World Limitations: What FSD Cannot Do

The vision-only architecture that He Xiaopeng praised is both Tesla's greatest strength and its most debated engineering choice. Cameras provide rich, high-resolution data that neural networks can process with remarkable nuance — reading text on signs, identifying pedestrian intent, and classifying objects. But cameras lack the depth-sensing certainty of lidar and the all-weather reliability of radar. In low-light conditions, heavy precipitation, or when camera lenses are obstructed by dirt or ice, the system's perception degrades faster than lidar-equipped competitors.

Real-world testing by multiple outlets has shown that city-street FSD requires driver interventions roughly every 10–30 miles in complex urban environments, though highway performance is considerably more robust — often going 100+ miles without a takeover on well-marked interstates.

How Tesla FSD Compares to XPeng XNGP and Other ADAS

He Xiaopeng's praise carries weight precisely because XPeng builds one of Tesla's most direct competitors in the ADAS space. XPeng's XNGP (XPeng Navigation Guided Pilot) is a city-and-highway capable system available on the G6, G9, P7+, and X9 models. Here's how the major systems compare:

ADAS System Comparison — 2026 Consumer Systems

System	Brand	SAE Level	Sensor Suite	City Streets	Cost
FSD (Supervised)	Tesla	Level 2	Cameras only (HW4)	Yes (US, expanding)	$8,000 / $99/mo
XNGP	XPeng	Level 2	Cameras + LiDAR (select models)	Yes (China primary)	Included on Max trims
Super Cruise	GM (Cadillac/Chevrolet)	Level 2	Cameras + LiDAR map + driver cam	No (highway only, mapped roads)	$2,500 or included on premium trims; $25/mo after trial
BlueCruise	Ford	Level 2	Cameras + radar + driver cam	No (hands-free highway only)	$2,100 / $75/mo after trial
Autopilot (basic)	Tesla	Level 2	Cameras only	No (highway lane centering)	Standard (included)

The key differentiators:

• Sensor philosophy: Tesla is camera-only; XPeng, GM, and Ford augment cameras with lidar and/or radar. Each approach has trade-offs in cost, redundancy, and weather performance.
• Geographic availability: Tesla FSD city-street capability is most broadly available across the US. XPeng's XNGP city features are primarily deployed in China, with limited international rollout. GM Super Cruise and Ford BlueCruise are restricted to pre-mapped divided highways in the US and Canada.
• Hands-free vs. hands-on: GM Super Cruise and Ford BlueCruise allow true hands-off driving on approved highways (eyes monitored via cabin camera). Tesla FSD still requires hands-on-wheel contact or torque, though its attention monitoring is increasingly camera-based.
• Pricing model: Tesla charges a premium ($8,000 or subscription). XPeng bundles XNGP into higher trims. GM and Ford offer trials then charge subscriptions.

Why He Xiaopeng's Praise Matters — and What It Signals

XPeng has invested heavily in its own autonomous-driving stack, deploying lidar on models like the G9 and P7+ to provide sensor redundancy that Tesla explicitly chose to eliminate. When He Xiaopeng praises Tesla's vision-only approach, it suggests several things about the state of the industry:

1. Neural-network maturity: Tesla's end-to-end neural network architecture — where a single model handles perception, planning, and control rather than modular rule-based systems — has demonstrated capability that competitors recognize as a significant engineering achievement.
2. Data scale advantage: Tesla's fleet of millions of vehicles provides an unmatched training-data pipeline. Every Tesla driving on roads worldwide contributes edge-case scenarios that refine the model.
3. Convergence possibility: XPeng has reportedly been exploring reduced-lidar or camera-dominant architectures for future vehicles, suggesting the industry may be converging toward Tesla's vision-first philosophy — albeit with selective sensor supplementation.

For consumers, this industry-level validation means Tesla's FSD stack is likely to continue receiving meaningful over-the-air improvements. But it does not change the fundamental reality: this is a Level 2 driver-assist system that requires your full attention.

Frequently Asked Questions