The highlight of FSD Beta v12 lies in its increased usage of neural networks. A neural network in artificial intelligence is akin to the human brain's network of neurons. It is a complex system that learns from large datasets - in Tesla's case, millions of video clips - to recognize patterns and make informed decisions. This machine-learning approach allows the system to adapt and improve continuously.
Neural networks are already used for determining surroundings, detecting objects, and more, but FSD v12 will be the first time Tesla adds neural networks that control the vehicle's acceleration, braking, and steering. According to Musk, FSD Beta v12 features end-to-end neural networks, meaning that these AI networks are used for everything, from vehicle input (cameras) to vehicle output (steering).
Tesla will be able to replace more than 300,000 lines of explicit code with these new networks.
Release Notes
Release notes for FSD Beta v12 were never obtained, however, the release notes for v12.1 confirm the use of neural networks for vehicle control.
The release notes are short, but hopefully, Tesla will expand them to offer additional details before v12 goes out to customers. The release notes currently only include one feature that states:
“FSD Beta v12 upgrades the city-streets driving stack to a single end-to-end neural network trained on millions of video clips, replacing over 300k lines of explicit C++ code.”
FSD Beta v12 Livestream
The public got an early glimpse of FSD v12's potential back in August 2023, when Musk live-streamed his vehicle driving with an alpha version of v12 on X.
The livestream wasn't without its share of issues, but it showed off some of its capabilities of v12 and it appeared to be at least on par with FSD Beta v11, even at this early stage.
Release Timeline and Inclusion of Updates
While the exact timeline for FSD Beta v12's release to public testers remains unknown, the expansion to additional employees is a promising sign. FSD Beta v12.1 is currently Tesla update 2023.44.30.10, although we'll likely see additional releases before it goes out to public testers.
Back in June of 2023, Musk said that FSD would finally graduate out of beta with FSD v12, however, the release notes for v12.1 clearly state "FSD Beta v12.1." Musk's goal could be for FSD to come out of beta with an eventual v12 release, but at this time it looks like FSD will remain in beta - at least in its current state.
FSD Beta v12.1 Video
It looks like we're not the only ones excited about the release of FSD Beta v12.
A video of FSD Beta v12.1 has been published online by an apparent employee. You can watch the video below.
Keep in mind that with major changes to FSD Beta, there are often regressions as well, and it usually takes several revisions to iron out major issues.
FSD Beta v12 is a fundamental shift in how the vehicle is controlled and represents an exciting future for Tesla and its FSD ambitions.
One of the big undocumented changes in Tesla’s 2024 Holiday Update was the changes to the Energy app. While the Model S, Model X, and Cybertruck received the Consumption tab in the Energy app for the first time, the changes made for those models also carried over to Model 3 and Model Y.
The Consumption tab lets you view your vehicle’s consumption over recent trips as well as view projected range estimates based on historical usage, but it now offers different options.
Sadly, legacy Model S and Model X vehicles produced before the 2021 refresh still don’t have access to the Energy app at this time.
Energy App
Tesla’s Energy App previously let you view a lot of in-vehicle data on what is consuming energy and how to improve your energy consumption. It was previously refreshed in 2022 and brought Drive, Park, and Consumption tabs to help compare actual vehicle energy consumption versus what you’d expect from the EPA ratings.
The old Energy App's consumption page.
Not a Tesla App
Key Changes
The Energy App has seen a lot of changes - mostly in the name of simplicity and reducing confusion. Some changes reduce functionality, but others bring even more. All of these changes impact the Consumption tab - the Park and Drive sections are unchanged.
Distance
Previously, you were able to switch the graph on the Consumption tab to show the last 5, 15, or 30 miles. Instead, it is now a static display of the last 200 miles (or 300km). This means your last 200 miles of driving - whether it's a single trip or multiple trips. Your range prediction and energy usage are now based on 200 miles of driving instead of the previous selectable distance.
This allows for a more reasonable range prediction as small bursts of high-energy usage, such as time spent accelerating to highway speeds from an offramp, are now less of an impact and are instead averaged out by regular driving.
However, for those who love to take their Teslas to the track or tow regularly, this makes the consumption significantly less useful because you can no longer see your actual energy usage for the type of driving you’re doing. This could be fixed with a reset button or by adding the ability to select your distance — similar to before.
Projected Range and Average Wh/mi
Unfortunatley, the Instant Range button has been removed, and the graph is now locked on what was previously the Average Range. Essentially, you cannot view your real-time range based on current instantaneous consumption - but you can view the overall projected range.
Additionally, average Wh/mi and projected range are still displayed - but in different areas compared to before. The projected range is displayed on the center-left side of the graph, while the average Wh/mi is now displayed at the top of the screen.
Not a Tesla App
Compare Vs EPA
Another new feature is that the average range is now compared to the EPA estimated range in terms of wh/mi. You’ll be able to see whether your driving style and conditions put you over or under the EPA estimate in a pretty quick way, which is helpful.
This new comparison is located just under your average Wh/mi.
Small and minor adjustments to your driving style - like not taking off like an electric lightning bolt at every red light - will make a big difference to your range. Don’t worry - we know its hard, we love doing it too! Other things - such as driving downhill versus uphill, will have an impact that you can’t necessarily avoid unless you’re old enough that you went to school uphill both ways.
Total Vehicle Consumption
The final new feature is a total vehicle consumption number at the bottom left, under the chart. It will tell you how much energy you’ve consumed over the distance you’ve driven so far. This is a convenient way of seeing exactly how much energy you’ve used.
Bug
There’s currently a bug in the way the Y-axis is labeled. The yellow area of the graph means the vehicle is using energy, while green means the vehicle is generating energy through regenerative braking. However, the Y-axis is currently mislabeled and shows generated energy as using about 100-200 Wh/mi.
The confusion appears to be due to the dark gray line, which looks like “0” on the Y-axis but actually represents the vehicle’s rated range. We would expect this to be Y-axis 0, since above the line the graph is yellow, and underneath it, the graph is green. However, this line is at about the 240 Wh/mi mark but will vary depending on the vehicle.
Due to this bug, it’s currently not possible to see how much energy is being generated.
Dynamic Y-Axis
The Y-axis in the Consumption tab is now dynamic - it will expand and contract automatically based on the driving data. We’ve seen it go from 400 Wh/mi all the way up to 800 Wh/mi. You likely need to be in a Model S Plaid or Cyberbeast with Launch Mode to see numbers much higher than that.
We’re sad to see the X-axis get locked to 200 miles, but being able to see total vehicle consumption and comparing average consumption against what’s rated is equally, if not even more valuable.
Overall, the new and improved Consumption tab is simpler and doesn’t require user input. While it takes away some features, it makes it easier for drivers who may not use it regularly. The most important piece is the projected range, which is now easier to see and understand unless you're towing and need the historical usage erased because it’s now irrelevant to your current drive. Hopefully, Tesla will allow you to scrub the graph horizontally in the future, adding the ability for the user to adjust the X-axis dynamically.
If you hop into your Tesla and say ‘Hi’ or ‘Hello’ after pressing the Voice Command button, there is a good chance it’ll reply with “Hello!” This is the newest and most interesting piece of news pointing us to the conclusion that a Tesla voice assistant is on the way.
Previously, if you tried this, it would simply return “Command not understood.” This is the first time the vehicle is responding and interacting with the user.
Experience It Yourself
You’ll need to have your vehicle language set to English. Once that’s done, you can use the voice command button on your steering wheel or yoke - for the Model 3 and Model Y, push the right wheel button, and for the Model S, Model X, and Cybertruck, press the button. Then go ahead and say Hi or Hello.
The Hello! response may even have regional differences. For a German Tesla owner, after setting his language to English, the response came back as “Hallo.” We’re interested to see what the responses may be in other regions, so let us know if you notice anything interesting.
We’ve tried a few other basic things, but it seems that, for now, the vehicle only replies to a simple greeting. Asking it what time it is or the $TSLA stock price doesn’t seem to do much yet - unless you’re in China with the updated Smart Assistant.
Not a Tesla App
Server-Side Update
This update appears to be happening over Tesla’s voice system backend and doesn’t require the Holiday Update. Users who aren’t on the Holiday Update are reporting that they’re getting this new response as well.
We already know that Tesla interprets speech remotely, and the driver’s voice is not processed in the vehicle. Instead, the voice snippet is transmitted to Tesla’s servers, where Tesla processes it and sends a response back to the vehicle so that the vehicle can interpret it. This is unlikely to change with a smart assistant, as Elon Musk has already said that Grok will still process data server-side instead of on-device.
Many users recently also noticed significant improvements to voice commands, saying that the system understands them better and that responses now come back faster.
All of these things point to a new backend system for voice processing that Tesla is testing. It’s not unusual for a company to switch to a new backend process but keep the capabilities the same as the legacy system until it’s ready to roll out the new features. At that point, it’s simply a flip of a switch to allow the new capabilities.
The new smart assistant that was rolled out in China is mostly a backend change, with the in-vehicle experience largely remaining the same. The activation method (button press) and user interface remain the same. What changed is the response that comes back from the server, and the assistant gained a voice. The new voice we receive with a smart assistant could very well be the new voice users are experiencing in the navigation system in newer vehicles.
Below is a video of the voice assistant in China:
Vehicle Support - Intel?
When China received the Smart Assistant, it was locked to cars equipped with AMD Ryzen processors only. Shortly after its initial launch, it became available to older cars with Intel Atom processors as well.
However, we’re not sure whether it would apply to legacy Model S and Model X owners. A legacy vehicle owner had their vehicle report “Command not understood” when they tried the ’Hi’ voice command.
Grok for Tesla
Elon has previously mentioned that Tesla vehicles would receive Grok AI. Grok, as of yet, still doesn’t have live speech support like other LLM models such as OpenAI’s ChatGPT or Google’s Gemini. However, a major update to Grok just brought massively improved image generation via a new model called Aurora.
xAI has been hard at work improving Grok, and we’re sure that live speech support is on its way soon. Once that feature arrives on X, Tesla will likely be well-positioned to enable a Grok-powered smart assistant fleet-wide with a flick of a switch.