FSD v14.3.2 Rolls Out With a Unified Model and Smarter Summon
Tesla pushed FSD v14.3.2 to owners this month, and the headline feature isn't the bug fixes. It's what the release notes call a "unified model" across Actually Smart Summon, FSD, and Robotaxi. That part is worth paying attention to.
What "Unified Model" Actually Means
The release notes state Tesla "unified the model between Actually Smart Summon, FSD, and Robotaxi for more capable and reliable behavior." Before this, those three systems were apparently running on separate neural networks, or at least separate training tracks. Now they share one. The idea is that lessons learned in one domain transfer to the others, so the system that learned to navigate a parking lot also benefits from what the Robotaxi fleet learned on open roads.
This could mean the gap between supervised FSD and Robotaxi performance narrows over time. One possibility is that Tesla has been intentionally merging these systems in preparation for the unsupervised FSD rollout that came up on the Q1 2026 earnings call. The announcement there was that unsupervised FSD would roll out gradually by geography, potentially state by state in the US. A unified model is exactly what you'd want before attempting something like that. Three parallel systems with different failure modes would be a lot harder to certify and deploy consistently.
Actually Smart Summon Is Actually Better Now
I've been skeptical of Actually Smart Summon since it launched. The "smart" in the name always felt like aspirational marketing. But v14.3.2 apparently addresses the core problem: the system now navigates more responsively and confidently to pinned locations. Confident is the word I care about here. Hesitant summon in a busy parking lot is worse than no summon at all, because at least with no summon you don't have strangers watching your car inch forward and stop four times before committing to a lane.
And combined with the parking improvements that came in v14.3, this update suggests the low-speed maneuvering side of FSD is getting sustained attention. That's the use case that matters in daily life for most people, more than highway performance.
The Double-Stop Fix
FSD v14.3.2 fixed double-stopping behavior at stop signs where road lines are present. If you've used FSD at intersections with painted stop bars, you know the issue: the car stops, creeps forward two feet, stops again. It's the kind of behavior that makes passengers nervous and makes you look like someone who just got their license. Glad it's gone.
The Disengagement Menu Is a Quiet Win
New in v14.3.2: when you take over from FSD, you can now log why. The categories are Critical, Comfort, Preference, and Other. This is useful feedback for Tesla and a decent self-tracking tool for owners. "Critical" is the category that matters most. If Tesla is collecting structured disengagement data at scale across thousands of cars, that's a much faster feedback loop than manually reviewing dashcam footage. It also gives the data some taxonomy, which raw counts never had.
What's Still Not Fixed
Region-specific signage is still causing inconsistencies. Specifically, "Except Right Turn" stop signs are tripping up v14.3.2. This is a genuinely hard problem. US signage isn't standardized across states, and training data reflects that unevenly. That's context, not an excuse. But it does explain why some of these edge cases take longer to close than straightforward bugs like the double-stop.
The Bigger Picture
The Q1 earnings call mention of a state-by-state unsupervised FSD rollout was light on specifics. No states named, no timeline beyond "gradual." But the unified model in v14.3.2 is the kind of technical groundwork you'd want in place first. A single system that can be tested and evaluated consistently is a prerequisite for any serious geographic expansion of unsupervised operation.
Whether that rollout actually happens on any particular schedule, I don't know. But the architecture is at least moving in the right direction.
Source: Teslarati