yep, I'm saying it already has the capability for full unsupervised FSD. What Tesla is doing, for now, is additional training and learning. Musk appears to be suggesting the crossover point, at least in terms of the capability, will come with FSD15. Maybe something talked about on the earnings call today. As of right now we are less than 300 million miles away from Musk's 10 billion mile target. Of note, less than two weeks ago we were just under 800 million miles away from that target. FSD is rapidly getting a lot of training data.

 

BMW and Mercedes Just Proved Tesla Was Right About Self Driving

https://www.autoblog.com/news/bmw-an...t-self-driving

Two German carmakers are quietly walking back their most ambitious autonomous driving bets.

By Simran Rastogi Apr 21, 2026

Key Points

• BMW and Mercedes halted Level 3 autonomy due to high costs and weak demand.
  
• Tesla avoided Level 3, relying on cameras over costly LiDAR, now seen as prescient.
  
• Infrastructure, regulations, and costs delay widespread Level 3/4 deployment until at least 2035.
  

BMW and Mercedes are stepping back from Level 3 autonomy. Both had bet heavily on eyes-off driving, but high costs, limited use cases, and weak demand killed the push.

In walking away, they've handed a quiet vindication to Tesla, the company the industry spent years mocking for refusing to go down the same road. Tesla held firm at Level 2+ and built its system around cameras rather than expensive LiDAR sensors. That last point drew particular ridicule. Cameras struggle in fog, heavy rain, and low-visibility conditions, but LiDAR does not. The consensus was that Tesla was cutting corners. It's now looking more like Tesla read the market correctly, and everyone else got ahead of themselves.

Why Level 3 Costs a Fortune

The SAE's autonomous driving scale runs from Level 0 (fully manual) to Level 5 (fully driverless). Most cars on the road today sit at Level 1 or 2, meaning the system assists but the driver remains responsible. Level 2+ systems, like Tesla's Full-Self Driving, can steer, brake, and accelerate independently but still demand constant human attention. Level 3 is the tipping point, where the car takes over, and the driver can legally disengage.

According to a McKinsey study, software development, testing, and validation cost four to seven times more at Level 3 than at lower autonomy levels. For BMW's 7 Series, the Level 3 option was priced at roughly $7,000 on top of an already expensive vehicle, and understandably, few buyers chose it.

A BMW spokesperson confirmed this, telling Automotive News that although the company brought the technology to production-ready status, no Level 3 function would feature in the revised 7 Series, because system costs and validation expenses remain very high.

Why the Roads Aren't Ready, and When They Might Be

The harder truth is that the world's infrastructure simply wasn't built for autonomous vehicles. Mercedes' Drive Pilot was initially capped at 60 kmph or 37 mph (later 95 mph or 60 mph), required a vehicle ahead of it to function, and only operated in clear weather on specific mapped motorways. Inconsistent lane markings, unpredictable road users, and fragmented regulations across countries make Level 3 autonomous driving difficult to guarantee.

The original promise of self-driving was compelling. Road accidents kill over 1.3 million people globally each year, and human error accounts for the vast majority. Autonomous driving could, eventually, change that.

But most experts now place meaningful mass-market Level 4 deployment somewhere between 2035 and 2040, and only in well-mapped, well-regulated urban environments. Not everyone is retreating in the meantime. General Motors is still pressing forward, testing an eyes-off, hands-free system in a fleet of 200 vehicles, with Level 3 targeted for its Cadillac Escalade IQ by 2028.

On the whole, the destination remains worthwhile. It's just that the road there looks increasingly like the one Tesla mapped out years ago.

 

If BMW was charging 7K for Level 3, it wasn't the cost that was the issue. It must have been the limitations.

 

Yesterday FSD made a small mistake, a mistake it really should not have made. I had FSD in Mad Max mode and there was a backup of cars in the right turn lane. For some reason, it changed lanes pulling out of the right turn queue to try and go around, but it really made no sense to do that because now it couldn't get back into the right turn lane. So it tried to make a right turn from the next lane over which wasn't dangerous, but it could 1) get you a ticket and 2) pissed people off who had to wait in the long line to make the right turn. I of course intervened. FSD tends to learn from these mishaps so it probably won't happen again at the same intersection, but issues like this (which are rare) should really not happen. Anyway, it was nothing dangerous, just annoying.

On another note, I have another coworker that may be getting a Tesla. Not because he thinks they drive awesome, not because they are super efficient or have an amazing 0 to 60, but because of FSD. After experiencing FSD in my Model 3, he was impressed. His daily commute is about an hour each way in traffic, and wants a Tesla simply for FSD. What's unique about this particular person is he is not an EV fan, and particularly never cared for Tesla. Being Middle Eastern, he grew up around gas guzzling Land Cruisers and Mercedes G Class SUV's. This technology impressed him so much that he's ready to trade in his E-Class for a Model Y

 

Took the grandsons to Disneyland this past weekend and FSD drove the whole way. From backing out of our driveway to parking up right in front of the hotel in Anaheim. Absolutely flawless, even using the feature for automatic HOV lane usage provided there are enough people in the car (and yes, it checks).

 

How long was your drive? What was the mix of dense/busy city streets vs. highway?

Interesting point about the HOV lane.👍 When I had a 2026 Model X, I had to "force" the car to get into the HOV lane (indicator stalk).

 

Around 90 miles door to door. Mostly highway of course, but busy. Few miles of country/urban to get from our house to the highway and a few miles of busy LA urban at the other end. The auto HOV is an option in Navigation. I'm assuming FSD took its cue from that as it took itself into the HOV lanes unprompted and it was clear it knew from the enhanced FSD visualizations that it was driving in an HOV lane.

 

The 5 from SD to OC (and back) on a weekend is no joke. We sometimes ride that stretch along Camp Pendleton on our road bike and the traffic is brutal. I can see how FSD makes that drive a little more bearable.

 

oh absolutely. I thought it made a massive difference.

 

Ah, that must have been the issue. Thx

 

I would tell you to replace your S with a 2026, but there is literally only 1 left in inventory

 

3 years from now I will get a 2026 Plaid S in mint condition with very low miles for about 50% of the new cost. 👍

Frost blue Plaid with a cream interior. 😁

 

Smart move 👍👍

 

but in 3 years you'll be old[er] life is not guaranteed.

 

Like you, @Hameed is young. Sure he can't escape death, but as a fellow Desi, I understand his motivation for bargains, even with death itself standing in the way