Raines (@raines1220)

Since v14.3, parking has regressed significantly for me because FSD keeps overriding my manual pin at home and work. It tries to “predict” a better parking spot and moves the P icon to random places. This might work for commercial lots like Walmart or Whole Foods, but it is terrible for my designated parking spot and my company’s parking garage entrance. Tesla, please don’t override manually saved pins. If I saved it manually, it is GOLDEN.

Waymo is much worse. From 2024-09 to 2025-12, according to Waymo itself (Safety Impact Data Hub), Waymo drove ~10.7M miles in Austin and had 86 accidents. That means they are at ~124.7K miles per accident, which is worse than the human baseline, and much worse than Tesla Robotaxi. And what’s worse? Waymo does not improve monthly like Tesla. When I say Tesla Robotaxi is ~1.81x safer than humans and ~3.62x safer than Waymo, that was from last month, and this month it will no longer represent the actual capability of Tesla Robotaxi. Also, don’t fall into Waymo’s trap when they compare “no-injury accidents” against humans and claim they are safer. That’s just a trick to make their numbers look better. I mean, Waymo drives like a grandma... it would be hard to have an injury at that speed and that level of hesitation.

Waymo is much worse. From 2024-09 to 2025-12, according to Waymo itself (Safety Impact Data Hub), Waymo drove ~10.7M miles in Austin and had 86 accidents. That means they are at ~124.7K miles per accident, which is worse than the human baseline, and much worse than Tesla Robotaxi. And what’s worse? Waymo does not improve monthly like Tesla. When I say Tesla Robotaxi is ~1.81x safer than humans and ~3.62x safer than Waymo, that was from last month, and this month it will no longer represent the actual capability of Tesla Robotaxi. Also, don’t fall into Waymo’s trap when they compare “no-injury accidents” against humans and claim they are safer. That’s just a trick to make their numbers look better. I mean, Waymo drives like a grandma... it would be hard to have an injury at that speed and that level of hesitation.

Tesla Robotaxi was at least ~1.81x safer than humans in April 2026. With the latest NHTSA crash report, and thanks to Robotaxi Tracker, I can finally report that Tesla Robotaxi in Austin surpassed average human safety in April 2026 (last month). 1. According to NHTSA, I estimate the human crash rate, including both reported and non-reported crashes, is ~249K miles per accident. 2. I collected all Tesla-reported mileage, as well as 7-day active fleet size from Robotaxi Tracker, and found the mileage is pretty predictable. This allows me to accurately estimate monthly Robotaxi mileage in Austin with <5% error. 3. Using the latest NHTSA crash report, I calculated the 3-month rolling crash rate, since Tesla sometimes has 0 monthly crashes. It clearly shows that in April, the 3-month rolling crash rate surpassed humans for the first time.

v14.3.2 is the first version where my reaction is no longer, “Wow, that feels very human-like.” Now it is more like, “Wow, a human could never do that.” Usually, I try to understand the “personality” of an FSD version. Is it chill? Is it too dramatic? Is it overly cautious? But v14.3.2 feels different. It almost has two personalities. In normal daily driving, it is smooth, natural, and human-like, almost like a friend driving you around. But when an emergency happens, it no longer feels like a friend. It becomes something else: a cold-blooded machine. Calm, smart, and fast. It handles dangerous situations correctly before a human even fully understands what is happening. That is why it feels trustworthy, but also kind of “scarily capable.” In certain aspects of driving, it is already more capable than even the best human drivers. And honestly, v14.3.2 is the first version that makes me seriously think FSD could save millions of lives that human drivers would never be able to save.

61.1% of Robotaxi rides in Austin over the past 7 days were unsupervised. The 30-day rate is 37.2%, and the 90-day rate is 17.9%. It is pretty clear that Austin Robotaxi will soon be fully unsupervised. When the next NHTSA crash report is released, I strongly suspect it will, for the first time, show “safer-than-human” statistics. It may not reach the level (5x–10x?) where you can confidently deploy unsupervised Robotaxi anywhere without a geofence, but it may be enough to show that it is slightly safer (~1.5x–3x) than human driving. If that is true, you know what that means? Since the model is unified across Robotaxi, FSD, and Summon, it is reasonable to assume that our “supervised FSD” has equivalent safety. That means even if we sit in the back seat, our car would still drive more safely than the average human.

Tesla shareholders funded the robotaxi vision through the hard part. The missed timelines, ridicule, drawdowns, and the “FSD is impossible” years. So if Tesla ever gets folded into SpaceX, the question isn’t “would that be cool?” It’s: who gets paid for the robotaxi value? Bradford breaks it down here.👇

What are the missing pieces for FSD? I think there are two of them: 1) a nice interface for a human to tell the vehicle their preferences (e.g., preferred parking spot), and 2) a really smart brain (e.g., reading text on signs). Do we need spatial measurements from LiDAR? Nope, it is a solved problem, I’m sorry.

First impression: FSD 14.3.2 is overall a more capable driver. It changes lanes more assertively, turns on signals at the perfect moment to prevent ambiguity, handles narrow spaces more accurately, and reacts to all potential hazards instantly at an earlier stage. I can easily tell that it drives safer than me, and I would feel safe even if you put me in the back seat. There is only one complaint: the P-icon parking spot prediction sucks. It overrides my previous manual pin and it is wrong. And it continuously predicts awful parking locations for different destinations.

Robotaxi Tracker has only 89 riders reporting their rides in Austin, and they’ve logged just 8,152 miles in total. Meanwhile, Austin Robotaxi has probably accumulated 1–2 million miles, meaning only about ~0.5% of the miles have been reported. From that ~0.5% of reported miles, 19 unique unsupervised Robotaxis have been discovered in Austin, and people are laughing about it. Well, to be honest, I’m laughing at you. Imagine I deploy 10 people to take Uber every day and count how many unique cars they discover. On the first day, they report discovering 10 cars. On the second day, they report another 10 cars (20 in total) Several days later, they report that 100 unique cars have been discovered in total (a pretty small number, right?). Then I can finally make fun of Uber: “Look! Uber only has 100 cars. How pathetic!”

I’ve seen everyone closely monitoring Robotaxi fleet growth. That’s a good thing, but keep in mind: 1. The “growth speed” is not the speed of Tesla adding a new car, but the speed at which someone reports a car that has already been added. 2. The fleet size does not represent all Robotaxis. It represents only the Robotaxis that have been reported by a limited number of volunteers.

Dan is the gold-standard quality inspector for the Tesla community. He used to be able to make videos about Teslas running over "kids". Now the best he can do is FSD 14.3.2 going “10 mph over the limit.”