News
Tesla FSD Beta 10.69.2.2 extending to 160k owners in US and Canada: Elon Musk
It appears that after several iterations and adjustments, FSD Beta 10.69 is ready to roll out to the greater FSD Beta program. Elon Musk mentioned the update on Twitter, with the CEO stating that v10.69.2.2. should extend to 160,000 owners in the United States and Canada.
Similar to his other announcements about the FSD Beta program, Musk’s comments were posted on Twitter. “FSD Beta 10.69.2.1 looks good, extending to 160k owners in US & Canada,” Musk wrote before correcting himself and clarifying that he was talking about FSD Beta 10.69.2.2, not v10.69.2.1.
While Elon Musk has a known tendency to be extremely optimistic about FSD Beta-related statements, his comments about v10.69.2.2 do reflect observations from some of the program’s longtime members. Veteran FSD Beta tester @WholeMarsBlog, who does not shy away from criticizing the system if it does not work well, noted that his takeovers with v10.69.2.2 have been marginal. Fellow FSD Beta tester @GailAlfarATX reported similar observations.
Tesla definitely seems to be pushing to release FSD to its fleet. Recent comments from Tesla’s Senior Director of Investor Relations Martin Viecha during an invite-only Goldman Sachs tech conference have hinted that the electric vehicle maker is on track to release “supervised” FSD around the end of the year. That’s around the same time as Elon Musk’s estimate for FSD’s wide release.
It should be noted, of course, that even if Tesla manages to release “supervised” FSD to consumers by the end of the year, the version of the advanced driver-assist system would still require drivers to pay attention to the road and follow proper driving practices. With a feature-complete “supervised” FSD, however, Teslas would be able to navigate on their own regardless of whether they are in the highway or in inner-city streets. And that, ultimately, is a feature that will be extremely hard to beat.
Following are the release notes of FSD Beta v10.69.2.2, as retrieved by NotaTeslaApp:
– Added a new “deep lane guidance” module to the Vector Lanes neural network which fuses features extracted from the video streams with coarse map data, i.e. lane counts and lane connectivities. This architecture achieves a 44% lower error rate on lane topology compared to the previous model, enabling smoother control before lanes and their connectivities becomes visually apparent. This provides a way to make every Autopilot drive as good as someone driving their own commute, yet in a sufficiently general way that adapts for road changes.
– Improved overall driving smoothness, without sacrificing latency, through better modeling of system and actuation latency in trajectory planning. Trajectory planner now independently accounts for latency from steering commands to actual steering actuation, as well as acceleration and brake commands to actuation. This results in a trajectory that is a more accurate model of how the vehicle would drive. This allows better downstream controller tracking and smoothness while also allowing a more accurate response during harsh maneuvers.
– Improved unprotected left turns with more appropriate speed profile when approaching and exiting median crossover regions, in the presence of high speed cross traffic (“Chuck Cook style” unprotected left turns). This was done by allowing optimisable initial jerk, to mimic the harsh pedal press by a human, when required to go in front of high speed objects. Also improved lateral profile approaching such safety regions to allow for better pose that aligns well for exiting the region. Finally, improved interaction with objects that are entering or waiting inside the median crossover region with better modeling of their future intent.
– Added control for arbitrary low-speed moving volumes from Occupancy Network. This also enables finer control for more precise object shapes that cannot be easily represented by a cuboid primitive. This required predicting velocity at every 3D voxel. We may now control for slow-moving UFOs.
– Upgraded Occupancy Network to use video instead of images from single time step. This temporal context allows the network to be robust to temporary occlusions and enables prediction of occupancy flow. Also, improved ground truth with semantics-driven outlier rejection, hard example mining, and increasing the dataset size by 2.4x.
– Upgraded to a new two-stage architecture to produce object kinematics (e.g. velocity, acceleration, yaw rate) where network compute is allocated O(objects) instead of O(space). This improved velocity estimates for far away crossing vehicles by 20%, while using one tenth of the compute.
– Increased smoothness for protected right turns by improving the association of traffic lights with slip lanes vs yield signs with slip lanes. This reduces false slowdowns when there are no relevant objects present and also improves yielding position when they are present.
– Reduced false slowdowns near crosswalks. This was done with improved understanding of pedestrian and bicyclist intent based on their motion.
– Improved geometry error of ego-relevant lanes by 34% and crossing lanes by 21% with a full Vector Lanes neural network update. Information bottlenecks in the network architecture were eliminated by increasing the size of the per-camera feature extractors, video modules, internals of the autoregressive decoder, and by adding a hard attention mechanism which greatly improved the fine position of lanes.
– Made speed profile more comfortable when creeping for visibility, to allow for smoother stops when protecting for potentially occluded objects.
– Improved recall of animals by 34% by doubling the size of the auto-labeled training set.
– Enabled creeping for visibility at any intersection where objects might cross ego’s path, regardless of presence of traffic controls.
– Improved accuracy of stopping position in critical scenarios with crossing objects, by allowing dynamic resolution in trajectory optimization to focus more on areas where finer control is essential.
– Increased recall of forking lanes by 36% by having topological tokens participate in the attention operations of the autoregressive decoder and by increasing the loss applied to fork tokens during training.
– Improved velocity error for pedestrians and bicyclists by 17%, especially when ego is making a turn, by improving the onboard trajectory estimation used as input to the neural network.
– Improved recall of object detection, eliminating 26% of missing detections for far away crossing vehicles by tuning the loss function used during training and improving label quality.
– Improved object future path prediction in scenarios with high yaw rate by incorporating yaw rate and lateral motion into the likelihood estimation. This helps with objects turning into or away from ego’s lane, especially in intersections or cut-in scenarios.
– Improved speed when entering highway by better handling of upcoming map speed changes, which increases the confidence of merging onto the highway.
– Reduced latency when starting from a stop by accounting for lead vehicle jerk.
– Enabled faster identification of red light runners by evaluating their current kinematic state against their expected braking profile.
Press the “Video Record” button on the top bar UI to share your feedback. When pressed, your vehicle’s external cameras will share a short VIN-associated Autopilot Snapshot with the Tesla engineering team to help make improvements to FSD. You will not be able to view the clip.
Don’t hesitate to contact us with news tips. Just send a message to simon@teslarati.com to give us a heads up.
Tesla has evidently revealed which of its vehicle models will be the first to receive Starlink integration: the Cybercab.
Tesla’s Santana Row showroom now has a full-fledged display of the Cybercab, with an extensive bit of information hung around an exhibit that seems to reveal the vehicle’s newest feature: an integrated Starlink antenna that will enable secure and reliable internet access during trips.
Credit: @Starscream_SJC | X
Cybercab is geared toward autonomous ride-hailing for one or two passengers. The production units rolling off the lines at Gigafactory Texas are built without steering wheels or pedals, meaning when public rides begin, passengers will not need to interact with a human being or control the vehicle in any way outside of what appears on the center screen for their entertainment during the ride.
Tesla Santana Row will be reopening tomorrow with a full focus on self-driving. Everything in the showroom is about Robotaxi and Cybercab with stats and information about the technology. The Cybercab on display is the production model. pic.twitter.com/yIUYdOGFOp
— Shaun Cassidy (@Starscream_SJC) July 20, 2026
Along the display, Tesla wrote this message about Cybercab:
“Cybercab is built for autonomy. It has no steering wheel, no side mirrors, and no pedals. It goes where you tell it to go and how you want it to, so you can relax along the way. It is hyper aware and responsive to your surroundings, monitoring other drivers, responding to emergency vehicles, utilizing its expertise in the rarest scenarios to help keep you safe.”
Tesla has been teasing a potential Starlink integration for quite some time now. In December, the company hinted at potential Starlink internet terminal integration within its vehicles in a patent that described a vehicle roof assembly with integrated radio frequency (RF) transparency.
The company wrote in its patent application that a new roof design built with materials that differ from the standard metallic or glass elements used in today’s cars would allow it to integrate modern vehicular technologies, in particular, ones that require radio frequency transmission and reception.
Tesla suggested high-strength polymer blends, like Polycarbonate, Acrylonitrile Butadiene Styrene, or Acrylonitrile Styrene Acrylate.
This is the first time we’ve seen Tesla officially confirm the Starlink integration into the Cybercab. It’s not much of a surprise considering the company’s intention behind the Cybercab, which is to make travel autonomous.
Productivity will now be at a maximum during a work-related commute, while the center screen could be utilized for Netflix or potentially even live TV for those who are heading to dinner or to a fun activity.
SpaceX has updated its target for the thirteenth integrated flight test of Starship, aiming for as early as Thursday, July 23. The 90-minute launch window opens at 5:45 p.m. CT from the company’s Starbase facility in South Texas.
The target flight was initially rescheduled for today, but SpaceX pushed it back again.
This latest adjustment follows an aborted attempt earlier in the week and reflects the iterative, rapid-development approach that has defined the Starship program. With the vehicle already stacked and ground teams making final preparations, the mission represents another step toward proving the full reusability of the world’s most powerful rocket system.
Now targeting to launch Starship’s thirteenth flight test as early as Thursday, July 23 → https://t.co/Rp7VwBzpWx pic.twitter.com/Y0YNzfc5zk
— SpaceX (@SpaceX) July 19, 2026
The original launch attempt on July 16 was scrubbed at T-0 when several Raptor engines on the Super Heavy booster failed to ignite properly. The automatic abort system triggered just as the engines began their startup sequence, preventing liftoff.
SpaceX CEO Elon Musk confirmed that some engines did not start as expected, prompting the decision to replace two Raptors on Booster 20 to ensure reliability. The issue occurred despite a successful full-duration static fire earlier, highlighting the complexities of coordinating 33 engines under flight conditions.
This cautious approach underscores SpaceX’s commitment to safety amid an aggressive test cadence.
Flight 13 builds directly on the lessons from Flight 12 in May 2026. The Super Heavy booster’s primary goals include a successful liftoff, ascent, stage separation, boostback burn, and controlled splashdown in the Gulf of America.
Hardware and software modifications address the off-nominal flip and boostback burn problems from the prior flight, where propellant slosh and engine relight issues led to an uncontrolled impact.
For the Starship upper stage, objectives include deploying 20 operational Starlink V3 satellites, the first real payload of this type, performing a single Raptor engine relight in space, and executing a controlled entry, descent, and splashdown in the Indian Ocean. Propulsion upgrades aim to improve engine-out capability after one vacuum Raptor was lost on Flight 12.
Additional test elements focus on heat shield performance. Six satellites carry cameras to image the tiles during flight, while white-painted tiles and upgraded attachments on flaps and the aft skirt will gather data for future reusability.
The FAA completed its mishap investigation into Flight 12 earlier this month, clearing the regulatory path.
This suborbital mission, the second with V3 vehicles, advances Starship toward operational missions, including potential crewed flights and support for NASA’s Artemis program. Success would mark significant progress in rapid reusability and satellite deployment from the massive system.
Elon Musk dismissed reports claiming SpaceX had placed a massive order for NVIDIA GPUs worth $52 billion. The denial came hours after Taiwanese media, citing unnamed industry sources, reported that SpaceX planned to acquire approximately 13,000 AI server racks, equating to roughly 1 million GB300 GPUs, from Foxconn.
Each rack was estimated at around $4 million, with deliveries potentially starting in late 2025.
The story suggested this would mark SpaceX’s first major foray into Foxconn-manufactured NVIDIA hardware, breaking from suppliers like Supermicro and Dell. Musk responded bluntly on X:
This is fake news
— Elon Musk (@elonmusk) July 20, 2026
Despite the denial, the rumored scale aligns with SpaceX’s explosive growth in AI infrastructure. NVIDIA’s GB300 (successor to the GB200 NVL) racks deliver unprecedented performance for large-scale training and inference. A $52 billion commitment would dwarf most corporate AI budgets and provide the compute muscle needed for frontier models.
SpaceX already operates gigawatt-scale terrestrial clusters like Colossus in Memphis, Tennessee, and has monetized them aggressively through leasing deals.
SpaceX’s newest Starmind will make earth data centers obsolete
Major customers include Anthropic (paying ~$1.25 billion monthly for 220,000+ GPUs), Google (~$920 million monthly for 110,000 GPUs), and Reflection AI. These arrangements are projected to generate tens of billions in annual revenue, far outpacing traditional SpaceX businesses.
Such an investment would fuel internal AI efforts, particularly Grok models under the integrated SpaceXAI division, while supporting ambitious orbital data center plans. SpaceX envisions launching thousands of AI-optimized satellites powered by solar energy and cooled in space, bypassing terrestrial power and land constraints.
This “Starmind” constellation could position the company as a leader in space-based computing.
SpaceX as an Emerging AI Powerhouse
Once primarily known for reusable rockets and Starlink satellite internet, SpaceX has transformed into a multifaceted AI player.
The 2026 acquisition of xAI integrated Grok development directly into the company. Starlink’s low-latency global network complements massive compute clusters, enabling efficient data flow for training and serving AI models.
Musk has long argued that AI scaling demands solutions beyond Earth, citing things like real estate and electricity limits on the ground.
While the Foxconn deal may not be in the cards, SpaceX’s trajectory is continuing on the path of blending aerospace engineering with hyperscale AI to dominate both launches and intelligence infrastructure.
Tesla reveals first vehicle model to receive Starlink integration
SpaceX adjusts Starship Flight 13 test launch target date once again
