Elon Musk announced today that Tesla “Full Self-Driving” v14.3 is currently in employee beta testing and will “probably go to wide release end of week.” Musk has described v14.3 as the version that will make your car “feel like it is sentient.”
If that sounds familiar, it should. Musk has promised that the “next update” will be the transformative one for about a decade now — and the most recent v14.2 updates have actually been degrading in key areas.
What Musk is promising
Musk posted on X today that “FSD 14.3 is in Tesla employee beta now and will probably go to wide release end of week.” This follows a March 19 confirmation that v14.3 was “in testing right now” with a “wide release in a few weeks.”
The CEO has been building up v14.3 as a landmark release since late 2025. Back in November, he called it the version where “the last big piece of the puzzle finally lands.” In a separate post, he claimed that “by V14.3, your car will feel like it is sentient.”
The update is expected to bring what Tesla describes as a significantly larger neural network with improved reasoning and reinforcement learning capabilities, better navigation routing, which has been a persistent complaint among FSD users, and enhanced handling of complex urban environments.
The update will target HW4-equipped vehicles first. HW3 owners, who make up the majority of Tesla vehicles that were sold the promise of full autonomy, will have to wait for a “FSD v14 Lite” that Tesla has vaguely targeted for sometime around mid-2026.
The v14.2 updates have been going backward
Here’s the problem. Tesla has been rolling out a series of v14.2 sub-updates over the past several months, and the results have been mixed at best. The latest v14.2.2.5 release has been described as “the most confusing release ever” because while it introduced some new capabilities, like school zone speed compliance and animal detection, it simultaneously introduced bizarre regressions.
Turn signal behavior became erratic, with the system activating right turn signals on sharp turns while continuing straight. FSD ignored navigation routing repeatedly, activating incorrect turn signals opposite to intended directions. The navigation system itself remains, in the words of one reviewer, “just terrible.”
And the early v14 release disappointed with hallucinations, brake stabbing, and speeding. Some data showed further v14 point improvements toward 2,000 miles between critical disengagement, but that went down closer to 1,000 miles with the latest updates and more data.
The regular disengagements in urban driving still happen every few dozen miles.
This is the reality of machine learning development: it is very much a two-steps-forward, one-step-back process. New capabilities come at the cost of regressions in previously stable behaviors. That’s normal — but it’s not what Musk is marketing.
A decade of ‘the next update will be the one’
It is difficult to place any trust in Musk’s claims about FSD timelines at this point. The track record is staggering:
In 2016, Musk promised a fully autonomous drive from Los Angeles to Times Square by end of 2017. In 2019, he said Tesla would have feature-complete FSD that year and that it would be safe enough to sleep in your car by 2020. In 2020, he was “extremely confident” of achieving Level 5 autonomy the next year. In 2022, he said he’d be “shocked” if Tesla didn’t achieve FSD safer than a human that year.
None of it happened. As we’ve covered, it’s the same pattern every single year: the current version is incredible, the next one will be transformative, and full autonomy is just around the corner. A court even ruled that Musk’s repeated self-driving predictions constitute “corporate puffery” — legally meaningless hype.
Now, “sentient” is the new buzzword. But FSD remains a Level 2 driver-assistance system that requires constant supervision, which is exactly what it was when Musk was promising unsupervised self-driving last year before quietly moving the goalpost to just “less nag.”
Meanwhile, NHTSA is closing in
The timing of this announcement is also worth noting. Just two weeks ago, NHTSA upgraded its investigation into FSD’s inability to handle reduced visibility conditions, sun glare, fog, airborne dust, to an Engineering Analysis covering 3.2 million vehicles. That’s the step that typically precedes a recall.
Top comment by HalfwitWizard
This would have been promising to hear 37 promises ago, but there's no consequences to failed predictions or outright lies. So what's the point of listening to him?
The agency found that FSD’s degradation detection system fails to warn drivers when cameras are blinded by common road conditions and that Tesla may be under-reporting related crashes. The probe has expanded from four crashes to nine incidents, including one fatality. This is now the third concurrent federal investigation into FSD, alongside a separate probe into traffic violations like running red lights.
Electrek’s Take
I know this is hard to hear if you’ve put any trust in Elon’s promises about FSD, but you need to keep the bigger picture in mind. FSD development is a two-steps-forward, one-step-back process, that’s just the nature of neural network-based systems. New training data improves one behavior while degrading another. It’s normal in machine learning, but it is fundamentally incompatible with Musk’s marketing that every new version is a breakthrough on the path to imminent autonomy.
V14.3 might be good. It might bring real improvements. FSD v14, broadly, has been the most impressive Level 2 system on the market. But “impressive Level 2 system” and “sentient” are not the same thing, and we are nowhere close to the unsupervised self-driving that Tesla sold to hundreds of thousands of customers for up to $15,000.
Musk has made this exact promise, that the next update changes everything, every year for a decade. The pattern is clear. Enjoy FSD for what it is: a capable driver-assistance feature that requires your full attention. Don’t buy the hype that v14.3 is about to turn your car into an autonomous vehicle. The data simply doesn’t support it.
FTC: We use income earning auto affiliate links. More.
Comments