Tesla has started rolling out FSD v12 to employees
Not a Tesla App
Elon Musk announced earlier this month that Tesla's Full Self-Driving (FSD) v12 would be released in two weeks. The usual timeframe reference Musk is famous for was met with skepticism. However, it seems that Tesla is right on track with its rollout.
We have learned through a trusted source that FSD v12 has started rolling out internally with Tesla update 2023.38.10.
Update: Musk has responded to our article on X, confirming that Tesla has indeed starting rolling out FSD v12 to employees.
FSD v12 is the update that is expected to remove "beta" from the title. The initial rollout to employees appears more limited in scale than previous updates. Considering the magnitude of the changes in this version, it makes sense to start slow.
Public Release
The timing of this internal release is close to two major Tesla events. The Cybertruck delivery event is just a few days away. Many eyes will be on the company during the event, allowing Tesla to possibly show the world its latest breakthrough. Alternatively, the highly anticipated holiday update, often regarded as the best update of the year, is expected to be released by 'Santa Musk' in the coming weeks, potentially featuring v12 as a significant addition.
Newer Features
The latest public FSD build, v11.4.7.3, is Tesla update 2023.27.7. This FSD build is several revisions behind the latest production builds, so it's nice to see that v12 will bring FSD beta testers back up to speed with some of the latest Tesla features such as Predictive Charger Availability, Faster Hazard Lights After a Crash, and other features included in updates 2023.32 and 2023.38.
As for FSD improvements, we haven't had a chance to see the release notes for FSD v12 yet. However, now that it has started going out to employees, it shouldn't be long before we find out all the FSD improvements included in this milestone release.
FSD v12: A Leap Forward with Neural Net Use
A significant change in v12 is eliminating over 300,000 lines of code previously governing FSD functions that controlled the vehicle, replaced by further reliance on neural networks. This transition means the system reduces its dependency on hard-coded programming. Instead, FSD v12 is using neural networks to control steering, acceleration, and braking for the first time. Up until now, neural networks have been limited to detecting objects and determining their attributes, but v12 will be the first time Tesla starts using neural networks for vehicle control.
The FSD v12 represents a significant leap in Tesla's FSD technology. Musk has described it as an "end-to-end AI", employing a "photon in, controls out" approach akin to human optical processing. This analogy underscores Tesla's ambition to replicate human-like decision-making capabilities in its vehicles.
Labeled as a "Baby AGI" (Artificial General Intelligence), the system is designed to perceive and understand the complexities of the real world. This philosophical and technological shift in AI-driven autonomy was vividly showcased during a live-streamed drive by Musk through Palo Alto, where the Model S demonstrated smooth and almost flawless navigation through various real-world scenarios, including construction zones, roundabouts, and traffic. That was three months ago; imagine how much the system has learned in 90 days.
Looking Forward: The Future of Tesla's AI Endeavors
The rollout of FSD v12 marks a critical point in Tesla's journey in AI and autonomous driving. It's not just about technological prowess but also about aligning AI with nuanced human behavior. With Musk's continued focus on AI, which is evident across his ventures, Tesla remains a crucial player in the EV market and the broader AI revolution.
As we await further details on the public release of FSD v12 and its potential showcase at the Cybertruck event, it's clear that Tesla is moving closer to a future where cars are not just self-driving but are also intelligent and responsive to the complexities of the real world.
Subscribe
Subscribe to our newsletter to stay up to date on the latest Tesla news, upcoming features and software updates.
Following yesterday’s news that the Tesla Robotaxi network is ahead of schedule and set to launch on June 12th, overnight, Elon confirmed that the Robotaxi network is far ahead of expectations.
With Elon back and working full-time at Tesla, SpaceX, and xAI, we’ll likely receive a lot more updates on what these companies are working on. In this case, Elon confirmed that Tesla has been running autonomous Model Ys with no one in the driver’s seat on Austin’s public seats, with no incidents. According to yesterday’s report, there is still someone in the passenger seat just in case something comes up, but essentially “driverless.”
Elon also confirmed that Tesla’s ambitious plans to have a Model Y deliver itself directly from the factory to a customer, presumably in Austin, will also occur next month.
For the past several days, Tesla has been testing self-driving Model Y cars (no one in driver’s seat) on Austin public streets with no incidents.
A month ahead of schedule.
Next month, first self-delivery from factory to customer.
While this will likely be geographically limited to the areas around Giga Texas, and potentially, Tesla’s Fremont factory in the future, it is still a great step toward reducing the cost of vehicles.
Public Access by End of June
In a reply, Elon also confirmed that anyone would be able to test out the Robotaxi network by the end of June. While in typical Elon fashion, he didn’t answer the question of whether the network will be invite-only at first, we expect this will likely be the case and it won’t just be anyone who can open the Tesla app and summon a robotaxi.
At the Q1 2025 Earnings Call, Elon stated that the Robotaxi service would only be available to the public in late June or early July, giving Tesla enough time to test with internal audiences, as well as invited testers. This timeline seems to be fairly concrete, but if you’re planning a trip to Austin to check out Robotaxi, we suggest looking at early July, as timelines can still shift.
Tesla’s Optimus X account recently dropped another video, following up on its previous dance routines. Previously, Optimus demonstrated impressive footwork, along with a level of agility and balance that has improved immensely since Optimus was first unveiled.
In a new video posted to X, Optimus is seen doing actual work. While it’s still in the early days, Tesla shows off how the robot is learning to perform small tasks.
The real focus of the video isn’t on the tasks Optimus is performing but on how it’s learning to perform these tasks. Optimus is learning by watching humans perform the work through a first-person view.
The video demonstrates Optimus tackling daily tasks that require a good degree of dexterity and environmental awareness, including vacuuming, cooking, and even taking out the trash. While the previous demonstrations were more of a showcase of Optimus’ physical capabilities, this one is more about the learning process.
The key breakthrough, which was emphasized by members of Tesla’s Optimus AI team, is the robot’s new ability to learn complex, bi-manual tasks largely from watching videos of humans performing them. This results in a reduction in direct teleoperation, where real humans control Optimus remotely to generate training data.
Murtaza Dalal, a member of Tesla’s Optimus AI team, shared his excitement about the development - and noted that while teleoperation doesn’t scale, training by watching video does scale — extremely fast.
First-Person Perspective
The ability to learn from human video (while currently focused on a first-person perspective) means Optimus can bootstrap new tasks much faster than before and with less human interaction required. Milan Kovac, an Optimus Engineering team member, elaborated on this, stating:
“We recently had a significant breakthrough along that journey, and can now transfer a big chunk of the learning directly from human videos to the bots... Many new skills are emerging through this process, are called for via natural language (voice/text), and are run by a single neural network on the bot (multi-tasking).”
What’s Next for Optimus?
The road doesn’t stop here. The Optimus team is already looking to see where they can push the barriers of learning, and is looking at having Optimus learn directly from internet videos of humans doing tasks. This means learning from third-person views rather than the more optimized first-person views that it is currently learning from today.
Once Optimus can translate the third-person perspective to the first-person, much like humans do, then it opens up a whole new world of training data. Optimus will be able to learn simply by watching the world around him.
All this rapid advancement is critical for Tesla. Elon has doubled down on Optimus being Tesla’s key to unlocking the biggest market in the world, and it is key to pushing Tesla’s new vision of sustainable abundance.
This is just the first step in a long process - we’re still years away from Optimus making its way to customer homes, but the journey has to start somewhere. For now, it’ll be working hard in Tesla’s labs and factories, and we expect to hear more about Optimus later this year once the production line is up and running.
![Tesla’s Optimus Is Now Learning by Watching — First Person Today, Third Person Tomorrow [VIDEO]](https://www.notateslaapp.com/img/containers/article_images/2025/optimus_cleaning.png/bd519f7f31d9c3e7e25c2736ee284efb/optimus_cleaning.jpg)
