For ten years, Elon Musk has promised the world that Tesla is on the verge of solving autonomy. Tesla still isn’t close. After years of failure, in 2023 Musk announced Full Self-Driving Version 12, which he promised to be the savior of Tesla’s woeful self-driving project.
When Musk announced that Tesla was developing Version 12, he claimed that it would represent a major advance in Tesla’s self-driving capabilities, as the new version of FSD would be “end-to-end AI”. As a result, FSD v12 would no longer “be beta” according to Musk. In a follow-up tweet, Elon said that v12 was “mind-blowing”, before claiming that it was “next-level” and a “breakthrough”.
  • But v12 was not mind blowing. After a year of development, v12 was shown to be just as bad as its predecessor, Version 11.
  • The number of miles between each critical disengagement for FSD v11.4.9 stood at 109, which was orders of magnitude away from driverless operation. The latest version of 12.5 is at 81 miles to safety critical disengagement.

Musk’s bold v12 pronouncements had followed a disastrous year of development for Tesla’s self-driving program. 2023 had seen Tesla’s flagship product falter, and Version 11 was widely seen to reflect the stagnation of Full Self-Driving’s progress, with even Tesla’s most ardent supporters calling for it to be removed from public roads.

Elon Musk and Tesla needed a radical departure from the current trajectory of FSD’s progress as it was becoming increasingly clear that devoting more time and resources into v11 was doomed.

Version 12 represented a new age in Full Self-Driving’s development, and was “an almost total rewrite” compared with previous versions. Tesla made a decision to discard years of work, removing the 300,000 lines of C++ code which had underpinned all previous versions of Full Self-Driving. The team that had failed for seven years to develop Full Self-Driving now pivoted to AI.

  • But this would fail as well. Tesla’s rewrite was for nothing. V12 was still a failure and Tesla would again fail to deliver on its promises.

Tesla hitched its wagon to the AI hype spreading through Silicon Valley, raising investors’ hopes that Tesla really should be considered a leader in the AI sector. The move was a major pivot designed to encourage investors to ignore Tesla’s stagnating auto sales amid increasing competition from other EV manufacturers.

Aligning with the new AI pivot, Elon tweeted that the successor to Hardware 3 and 4 would be “renamed to AI5”, which supposedly would offer ten times the capabilities of HW4.

  • But Hardware 4 had been no better than Hardware 3. The latest version of FSD running on Hardware 4 still only traveled 109 miles between safety critical disengagements. There is no reason to believe Musk’s promises about AI5.

Tesla started again after nearly a decade of futile efforts to solve autonomy. Musk’s hubris led him to arrogantly claim that Tesla could start from scratch and within a year leapfrog Google’s Waymo, which had been perfecting fully autonomous driving since 2013.

Tesla’s pivot to AI allowed Elon to distance himself from some of the investor impatience associated with his years of broken promises about achieving autonomy. The rupture with previous versions of Full Self-Driving by completely rewriting the code base of FSD could buy Elon valuable time to convince investors that Tesla’s new approach would soon yield results.

  • But within a year of this pivot, it was clear that Tesla’s AI approach with v12 was an abject failure.

Musk was aware that a drastic improvement in miles-per-intervention was urgently needed to show investors and regulators that Tesla was in sight of autonomy. Without any measurable proof of success, regulators would never approve a Tesla robotaxi on public roads. He baselessly claimed that FSD improvements would come in more rapid succession than ever before.

In March 2024, he claimed that FSD would receive regular updates and that the software would be “shining bright” in just a matter of months.

  • But in September 2024 FSD v12 is still a failure. In fact, v12.4 was so woeful that Tesla was forced to scrap it before it was ever sent to wide release.
  • Tesla did not roll out any new improvements every two weeks. These turned out to be minor versions with a month between them.
  • They were not significant improvements, as demonstrated by the disengagement data. By late April and early May, v12 was a disaster.
  • None of Musk’s objectives had been met, and, in fact, a number of FSD “testers” were experiencing regressions with the new build.

Elon claimed in June that v12 would finally see the introduction of “Actually Smart Summon (ASS)”, enabling consumers to summon their car with computer vision with no-one in the driver’s seat in “a month or two”, before stating in July that the feature would be available “next month”.

  • Despite Musk claiming in 2016 that a Tesla owner could summon their car from the other side of the US by 2018, the release version of ASS is limited to 200 feet.
  • ASS is not even legally allowed to operate on public roads. Elon had yet again made a promise which Tesla is still to this day unable to deliver.

Another key component of the v12 stack was that Cybertruck owners would be able to use Full Self-Driving for the first time. In May, Cybertruck owners were told that Full Self-Driving was only “a few months away” and in mid-July Elon tweeted that Cybertruck FSD was imminent.

  • However, when FSD v12.5 was eventually rolled-out in August, Full Self-Driving did not become available for Cybertruck owners, and as of September 2024, it has still not been released. Those who have paid for the Full Self-Driving package, which is included in the price of the Foundation Series Cybertruck are still waiting.

Musk also promised that FSD v12.4 would achieve “roughly 5X to 10X improvement in miles per intervention vs 12.3”:

Musk then claimed v12.5 was “a major improvement in” terms of miles-per-intervention compared with terrible v12.4. In May, he had tweeted that 12.5 would focus on fine-tuning FSD, and address “​​rare, complex situations, for example: going down a narrow, one-way road, encountering a road closure and having to reverse out to find a new route”.

  • Again, FSD 12.5 was not a “whole other level” as Musk had described. v12.5 travels 109 miles between safety critical disengagements while v12.3 travels 112.
  • Users complained that FSD 12.5 almost rear-ended another vehicle, while others observed that FSD 12.5.1.3 failed “to yield for a bus with flashing red and stop sign out”. Tesla has repeatedly failed to deliver a safe and non-defective update which was any better than the ones which came before it.
  • Musk’s claims about FSD’s miles per intervention improving were a total lie/

The evidence from third-parties demonstrates that FSD 12.5 still contains a concerning array of safety critical defects.

Despite evidence that FSD v12 was regressing in many areas, Elon doubled down and said that Tesla would double the average number of miles FSD could travel between interventions in August.

This also failed. The number of miles between interventions actually dropped by 50% since Elon said this, proving Elon’s promises to be false yet again.

Despite all of these repeated failures, Tesla’s AI roadmap says that v12.5.2 will see a 3x improvement on miles between necessary interventions – a target which will inevitably not be met.

Having failed to meet his promise of 5x to 10x improvements in disengagement rates with v12.4, and then failing to double average miles between interventions in August, Elon then said it would “take over a year of driving to get even one intervention” on v12.5.

  • He failed again. Version 12.5 can only travel 109 miles between safety-critical disengagements.
  • After each failure to meet these ridiculous targets, Musk desperately doubled down again and promised even more lofty goals, recently announcing that v13 will be released in October 2024, with a 6x improvement on miles between necessary interventions.

Elon’s promise that FSD 12.5 would finally merge the city streets and highway stacks of Full Self-Driving has also failed to materialize. In early August, Elon stated that the merging of stacks would occur “later this month”, in conjunction with the release of v12.5 on Cybertruck.

  • This has not happened.

Hardware 3 owners have fallen victim to Elon’s recent false promises about FSD v12, after they experienced severe delays to receiving FSD 12.5. Elon Musk and Tesla promised since 2016 that HW3 vehicles would be capable of becoming robotaxis. In 2019 Musk said there would be 1 million Tesla robotaxis on public roads and that their owners would earn $30,000 a year from them.

  • Tesla has never deployed a single robotaxi.
  • No Tesla owner has ever made a single penny from their car operating as a robotaxi.

Elon stated in June that HW3 owners would receive the latest version of FSD in “10 days”.

  • This was another broken promise. Most HW3 vehicles have not received the update. A select few owners only began downloading 12.5 on their Hardware vehicles in late August.

Tesla’s planned robotaxi unveiling event was delayed from its original August date to October.

  • The delayed event showed just how badly Musk had failed.
  • It was clear that Tesla could not improve the number of miles-per-intervention on v12 by any meaningful degree, despite Musk’s desperate promises.
  • Tesla cannot even hope to achieve the same poor level of performance on HW3 vehicles without significant effort and expense.

Recently, there have been indications that Tesla has been frantically training its vehicles on data around the Warner Bros. Studios where the unveiling event is set to be taking place in October.

  • With only a month to go, it is clear that Tesla has still not developed a version of FSD which can be considered to be a ‘Robotaxi’.
  • Tesla is willingly misleading customers by training its data on a small space to impress investors and give an indication that it performs better than it actually does.
  • In October 2016 Tesla even conducted a staged unveiling of its Solar Roof project at another Hollywood Studio. As detailed by Walter Isaacson, this demonstration was a total fake and none of the “solar” roof tiles shown were functional solar panels.

The Full Self-Driving roadmap that Tesla published in September 2024 shows just how desperate Elon now is.

Tesla is now claiming again that version 12.5.2 of FSD will improve miles per intervention 3x before September is over.

  • This will not happen. Tesla says 12.5.2 will run on the AI3 computer, but has to date failed to implement this for 12.5.1.

In October, Musk and Tesla say they will kick v12 to the curb in favor of the shiny new v13, which they claim has 6x improved miles per intervention.

  • It is ridiculous that Tesla now claims they will achieve this, after failing on every attempt with v12.
  • Every prediction Musk has made about FSD v12’s improvements in miles per intervention has been false. v12.4 was not a “5x to 10x improvement in miles per intervention” compared to 12.3. Its rollout was stopped due to how woefully defective it was.
  • There is no reason to believe that v13 will be 6x safer. If Musk had been right about v12’s 5x to 10x improvement, Tesla would have been getting somewhere with FSD, but he was wrong, and the technology is still as terrible as ever.

Tesla’s roadmap outlines an upcoming ability to “unpark, park and reverse in FSD”, scheduled for October.

  • It is woefully defective compared to other automakers’ systems. There is no reason to believe that it will improve in October.
  • In 2016 Musk claimed in a tweet promoting Tesla’s staged Full Self-Driving demonstration that a Tesla could park itself autonomously, but it could not then, and still cannot to this day.
  • In March 2020, he said that it was “not long now” until reverse summon would be available. Nearly 8 years later, a Tesla still cannot park autonomously, and requires human intervention to avoid making dangerous mistakes.

Tesla now says that it will deploy its defective self-driving AI internationally, in the EU and China.

  • But only eighteen months ago, Musk said that Tesla was adapting FSD version 11.3 to European roads. This was a total failure and v11 was consigned to the dustbin of history, with no international version launched. There is no reason to believe v12 will be any different, given its litany of critical safety defects.

To make matters worse, Tesla’s roadmap gives no indication of when it actually expects to solve autonomy.

Musk proclaimed that he could solve a problem which no-one had ever previously solved in just one year. This was underpinned by the same hubris that caused Musk to say in 2016 that autonomous driving was a “solved problem”.

Nobody should believe a word that Elon Musk has to say about Tesla Full Self-Driving, as shown by his repeated lies about v12’s capabilities. Tesla’s defective self-driving software is doomed to fail.

The robotaxi market is already saturated and Waymo is growing exponentially, and will soon be in the majority of metropolitan areas in the US. In China, Tesla’s second largest sales market, robotaxi companies such as Apollo Go, Pony.ai, WeRide, AutoX and SAIC Motor already dominate the robotaxi industry.

Tesla is still unable to make FSD work despite a total rewrite. It is clear that they are unable to solve autonomy. Tesla’s robotaxi business is dead on arrival.