Tesla’s Full Self-Driving (FSD) software has been a subject of immense curiosity, excitement, and sometimes skepticism. Yet, the progress Tesla has made in developing and refining FSD is nothing short of remarkable. During a panel discussion on Hans Nelson’s YouTube channel titled Tesla’s Insane FSD Progress, Tesla experts and investors Hans Nelson, Herbert Ong, and Jeff Lutz discussed the latest FSD advancements, shedding light on the exponential improvements Tesla has achieved and the challenges that lie ahead.

“Tesla has a really good ability to sort real-time issues as they come in from drivers,” said Herbert Ong. He praised the company’s data-driven approach, which has been instrumental in refining FSD at an unprecedented pace. With version 12.5.2.1 already deployed to 50,000 vehicles, Ong explained that Tesla’s responsiveness to user feedback has made their software continually better. “The progress is really good, and it’s interesting to see how quickly they’ve addressed things like abrupt stopping at yellow lights or hesitation at green lights—all things that drivers have been reporting,” Ong added.

Listen to our conversation on the exponential improvement of Tesla FSD!

Recent Updates and What’s Next

One of the most notable milestones discussed during the panel was the full rollout of end-to-end highway driving to all AI Hardware 4 users, which is targeted for early next week. “They’ve enhanced stop smoothness, reduced the annoying weather notifications, and introduced various safety improvements,” said Jeff Lutz. Version 12.5.x continues to refine city driving, and Smart Summon has now been released to Europe, China, and other international markets—a testament to Tesla’s ambition to make their technology globally adaptable.

However, Ong also touched on the elephant in the room: “There are some features that were planned for October, like the unpark, park, and reverse capabilities, which haven’t yet been fully rolled out.” He emphasized that although Tesla fell behind on these particular features, the sheer scale of the We Robot event, which took place in mid-October, might have been the reason for the delay. Despite some setbacks, Ong remarked, “Even if all of this gets pushed to December or January, the bigger picture is that these engineering problems have solutions, and Tesla will deliver.”

Harnessing Data to Drive Improvement

The role of data cannot be overstated when it comes to Tesla’s FSD improvements. “The amount of information they’re getting from the fleet is pretty significant,” Lutz said. “We’re talking about roughly 50,000 vehicles on version 12.5.6, and when you have datasets this large, it becomes much easier to identify product performance and trends in real-time.” He explained that analyzing large data sets is crucial for understanding how the system is performing and where it can be improved. “When you’re dealing with millions of data points, you get a much clearer picture compared to smaller, more discrete datasets,” Lutz added.

This approach to iterative testing and data collection enables Tesla to identify issues quickly and adjust course where necessary, which has been a driving factor in their progress. “Tesla’s ability to validate these changes through large-scale data analysis really sets them apart,” noted Ong.

Hardware 4 and AI4: Unleashing New Capabilities

Tesla’s Hardware 4, coupled with the latest AI4 models, represents a significant leap forward in the company’s self-driving capabilities. “The AI4 hardware has not only a faster computer but also higher resolution cameras,” said Nelson. “Up until now, they’ve only been emulating Hardware 3 cameras, but now they’re taking the training wheels off and running AI4 in full native resolution.” This upgrade enables Tesla to fully leverage the compute power and higher-resolution imagery that AI4 offers over previous hardware versions.

Nelson also addressed a common concern among Tesla owners about Hardware 3’s ability to keep up with future software updates. “Management has mentioned that while they are developing at the cutting edge with AI4, they will eventually backport these improvements to Hardware 3,” Nelson explained. “There are millions of Hardware 3 vehicles on the road, which makes it worthwhile to compress the models and adapt them for the older platform.” He added that, if necessary, Tesla could retrofit the older vehicles with upgraded computers—a complex but feasible solution.

Improvements in Version 13: The Next Leap

Version 13 of FSD is poised to make substantial improvements across the board. “They’ve already said that version 13 has a four-times increase in miles between necessary interventions compared to version 12.5.4,” Lutz pointed out. This version is expected to be delivered to internal customers imminently, with a wider rollout targeted around Thanksgiving. “Much like version 12, which took a while to reach a full rollout, version 13 is a huge leap and will be phased in over time,” said Nelson.

The panelists detailed several of the new features included in version 13, such as improved collision avoidance systems, enhanced traffic control responses, and even the ability to detect emergency vehicles through audio inputs. “One user actually showed that the car could see an emergency vehicle coming and moved to the shoulder—an impressive feature that speaks to the depth of the FSD system,” Nelson said.

These advancements also include significant improvements in AI architecture. “They’re scaling everything—three times the model size, 4.2 times the model context length, and five times the training compute,” Lutz explained. “It’s clear they’re throwing everything at improving FSD, using the much-expanded version of Cortex. This means bigger, better, and more advanced models that will continue to improve the performance of FSD.”

Challenges of Scheduling and Development

Despite the rapid pace of improvement, there are inherent challenges to developing a technology as complex as Tesla’s FSD. “This isn’t like upgrading iOS from version 16 to 17,” Lutz said. “We’re talking about entirely new capabilities that have never been done before, so the ability to schedule down to the day becomes extremely difficult.” He emphasized that Tesla is making progress at an impressive pace, especially when compared to previous versions of FSD. “They’re now converging on more predictable timelines, and the delay has narrowed to about a 30-day window, which is pretty impressive given the complexity,” Lutz added.

A Global Push: Expanding FSD Beyond the U.S.

Another topic of interest during the panel was Tesla’s expansion of FSD capabilities to international markets. “Smart Summon has now been released globally, including Europe and China, even though FSD supervision isn’t yet fully approved in these regions,” said Ong. This move, he suggested, signals Tesla’s confidence that regulatory approvals are not far behind. “It’s easier to roll out these features in other regions once you already have them functioning elsewhere,” Nelson added.

The Exponential Path Ahead

The panelists were unanimous in their belief that Tesla’s Full Self-Driving system is improving at an exponential rate. “They’re working through a punch list of engineering problems, and they’re getting it done,” said Nelson. “The continued upgrades in hardware, AI models, and the sheer amount of data they’re processing means that FSD is on an upward trajectory that’s going to be very difficult for competitors to match.”

In the near future, Tesla’s FSD system will be increasingly capable, reliable, and ubiquitous. As Ong concluded, “Even if there are delays, Tesla’s track record shows they will deliver. These are engineering problems with solutions, and Tesla is one of the few companies that have both the technology and the ambition to solve them.”

With the combined power of advanced AI, robust data collection, and continued investment in hardware, Tesla’s Full Self-Driving is poised to redefine what’s possible in autonomous driving—not just in the U.S., but globally. The road ahead may still have bumps, but it’s clear that Tesla is accelerating at full speed toward a future where cars truly drive themselves.