Skip to content 
Tesla’s Full Self-Driving (FSD) v12.3.6 takes a bold step forward in the race toward autonomous mobility. In a recent real-world test conducted by the Munro team, this supervised autonomy software was pushed through a diverse set of urban and expressway scenarios. The result? A mostly smooth, intuitive, and confidence-inspiring ride that raises the bar for consumer-grade self-driving systems.
In this deep-dive, we unpack what happened during the test and assess where FSD v12.3.6 shines—and where it still has room to grow. Whether you’re an EV enthusiast, automotive engineer, or investor tracking autonomous tech, this road test offers critical insights into the present and future of Tesla’s approach.
Navigating Urban Terrain with Minimal Supervision
Upon initiating Tesla’s full self-driving supervised mode through the vehicle’s autopilot menu, the software version 12.3.6 was confirmed. The journey began with a set urban route designed to test FSD’s proficiency in stop-and-go conditions, lane discipline, and merging decisions.
Immediately, the system exhibited intelligent but occasionally debatable behavior. For instance, it opted for a left lane when the driver intended to make a right turn. While this was not incorrect from a rule-based perspective, it revealed the algorithm’s preference for structured interpretations over context-sensitive decisions.
A consistent strength was FSD’s adherence to road markings. It reliably stopped at bold white lines at intersections, demonstrating disciplined spatial awareness. However, in one poorly marked intersection, the vehicle veered slightly off the intended lane before self-correcting—a minor miss that could escalate in denser traffic.
Highway Handling: Expressway Merging and Exiting
Once on the expressway, FSD v12.3.6 revealed its most compelling competencies. Merging behavior was efficient, with the car ramping up to speed and safely entering traffic. Throughout the expressway journey, the driver did not need to intervene—a remarkable sign of reliability at higher velocities.
The system also demonstrated anticipatory behavior near exits. It consistently initiated lane changes to prepare for exits at around 0.7 miles from the junction, offering a seamless transition. These observations suggest the software’s path planning and predictive modules are maturing rapidly.
Another highlight was its treatment of large vehicles like semi-trucks. As the Tesla approached or passed these vehicles, the system intelligently biased slightly within the lane to provide added clearance—mimicking cautious human behavior and enhancing perceived safety.
Roundabout Readiness: A Critical Benchmark
Roundabouts are notoriously complex even for human drivers. FSD tackled multiple roundabouts in this test, showcasing a range of behaviors. It preferred the outside lane—an approach that worked effectively when the route required driving straight through but would need modification for more complex exits.
One notable moment came when the system needed to navigate entirely around a circular junction to return on a different road. Impressively, the vehicle completed the maneuver with minimal hesitation, adhering to proper flow and signaling conventions. This is a significant improvement over earlier FSD versions, which often struggled with roundabout logic.
Nonetheless, there was occasional ambiguity in lane selection entering the roundabouts. The vehicle consistently chose a path, but the logic behind the choice wasn’t always clear—possibly indicating reliance on GPS data over visual context.
Stoplights, Lane Endings, and Parking Precision
In city conditions, the software’s decision-making continued to impress. For example, at a “no turn on red” intersection, it waited patiently for a green arrow. Moments later, at another junction where a right on red was legal, the vehicle correctly executed the turn—despite pressure from an impatient driver behind.
It also correctly responded to changing road conditions such as a lane ending. Rather than hesitating, it assessed the moment and merged smoothly—another sign of enhanced situational awareness.
Parking performance was another standout feature. Upon arriving at the destination, FSD v12.3.6 identified a viable parking space, reversed in cleanly, and parked with impressive speed and precision. The user remarked that the system parked “better than I do,” emphasizing the practicality of this automated feature.
Construction Zones and Edge-Case Decisions
While the software excelled in most conditions, certain edge cases still required human oversight. When approaching a construction vehicle with directional arrows, the driver chose to manually activate the turn indicator out of caution. Although the system likely would have handled the situation correctly, the driver’s proactive engagement highlights that trust in FSD still has bounds.
Interestingly, the system allows for manual turn signal input without overriding the entire autonomous control. This cooperation between human and machine reflects a hybrid mode of operation that may become the norm before full autonomy.
Summary: A Confident Step Toward Autonomy
Tesla’s FSD v12.3.6 performs exceptionally well in a wide range of real-world conditions. From urban stoplights to multi-lane expressways and complex roundabouts, the system displays robust decision-making, contextual awareness, and smooth execution. While it still benefits from human supervision in ambiguous situations, it increasingly demonstrates characteristics of a mature, road-ready autonomy suite.
Key Takeaways:
- Expressway performance: Smooth merges, exits, and intelligent vehicle spacing
- Urban handling: Accurate stops, legal right-on-red decisions, and lane management
- Roundabouts: Navigated successfully, though lane logic remains opaque
- Parking: Precise and efficient, with minimal adjustment needed
- Limitations: Occasional lane selection misjudgments and uncertainty in edge cases
Final Verdict
FSD v12.3.6 isn’t perfect—but it doesn’t need to be. It brings Tesla closer than ever to delivering on its promise of practical, scalable autonomy. For drivers willing to stay engaged while the system handles 90% of the task load, it offers a glimpse into the future of driving.
Tesla’s approach still emphasizes human oversight. But as this test reveals, that oversight is becoming less about control and more about confirmation. With software improvements accelerating and real-world validation ongoing, FSD v12.3.6 might soon become the benchmark by which all other semi-autonomous systems are measured.
Explore More Autonomous Tech with Munro
Dive into more teardown insights, cost breakdowns, and real-world EV tech reviews at Munro & Associates. Or check out Munro Live to see how other advanced driver-assistance systems stack up.