Volvo did something carmakers almost never do. One model year into the EX90, it swapped the primary drive axle end for end, raised the pack voltage, changed the electrical architecture, deleted the lidar, and left the badge alone. Out of Spec Reviews calls the result SPA2 Upgraded, and what it means for whoever bought the 2025 version is: nothing. No over-the-air update crosses that line. Two cars, one name, and a used listing that will not tell you which is which.
Once you notice the line, it runs through most of the day. Kim Java covers an owner who physically moved a 2022 Model 3 from HW3 to AI4 using parts from a 2026 Model Y, and the finding that matters is not that it booted. It is that all eight cameras had to come along, and that the full self-driving purchase stayed behind with the old computer. It also reports vehicle to load arriving for the Model Y and skipping the early build dates, with the retrofit quoted at roughly $1,700 to $1,800 in parts and labor. Rivian, per the same episode, has told buyers plainly that the lidar-equipped Gen 3 hardware is the one that eventually goes eyes off, and that Gen 2 cars have about two years of parity before that gap opens.
What software can reach is the other half of the day, and it is not trivial. David Vaught spent an afternoon cycling an R2 through three named aggression settings, and by his own rough count the highway following distance goes from about ten car lengths in mild to under three in spicy. Same sensors, same computer, three different drivers. Edmunds makes the ownership version of the point: its long-term R1T gained range and ride quality from updates over a year. Software moves the product. It just cannot move a camera's resolution or a cell count.
Out of Spec Dave's recovery run is the reminder that the line also sits where nobody documents it. Four EVs were driven until they stopped, and each behaved differently at the bottom: the Model Y held 70 mph until it quit, the R2 took a charge straightaway, and the R1S dropped its 12 volt system hard enough, on camera, to need a reset before it would accept power. None of that is on a spec sheet.
The thing to watch is whether any of this becomes purchasable. A retrofit finished in a driveway is an existence proof, not a service offering. Tesla has said it intends to for HW3 owners who paid for FSD, but no price has surfaced. Until one does, the oldest advice in the EX90 video holds: buy what the car does on the test drive, not what it might do later.
Volvo Rebuilt the EX90 After One Model Year and Kept the Name
Out of Spec Reviews has a model year 2026 EX90 for a two week test and opens by saying it is not a facelift. Kyle reports the platform is now called SPA2 Upgraded, with a new powertrain, battery, electrical architecture and no lidar. The first car ran a front primary axle with clutch disconnects; this one is rear primary, permanent magnet at the back and induction up front. He puts the gain at roughly 130 horsepower for about 680 total, with pack capacity down from around 111 kWh to around 107.
Charging is the trade he thinks is worth making. The old car peaked at 250 kW but needed north of 600 amps to get there, which he says US stations rarely supply, noting Tesla caps non-Tesla vehicles at 500 amps. The new car runs higher voltage, and Volvo claims 350 kW and 10 to 80 percent in 22 minutes, untested so far.
His objections are specific. The car keeps a CCS port despite being built in South Carolina while the EX60 gets NACS, there is no dog mode, and he says the seat backrest makes the same noise he flagged two years ago. He also reported laggy screen brightness and a jolty accelerator. On the 2025 cars, his understanding is that the Luminar relationship dissolved and the lidar will never be switched on, and that Volvo is compensating owners in Norway, though he is not certain about the US.
An Owner Moved a Model 3 to AI4, and the FSD Purchase Did Not Come With It
PJ and Kim report that a Tesla owner and software engineer pulled the HW3 computer out of a 2022 Model 3 and fitted AI4 hardware from a 2026 Model Y. The wiring stayed put, with adapters bridging old harness to new, and no cutting or drilling. All eight cameras had to be replaced, moving from 1.2 megapixels to as much as 5. He estimates $4,000 to $6,000 in parts, and error messages appeared on screen during the process.
Two catches sit behind the headline. They point out a warranty would not survive it, and the full self-driving entitlement stayed with the old computer, so the car needed a fresh subscription. Anyone who paid outright for FSD on HW3 would be buying it twice. Kim Java's read is that this matters less as a DIY project than as evidence Tesla could offer the swap as a service, which the episode says Tesla has committed to for HW3 owners who bought FSD.
The same episode has the cleaner example of a build-date cutoff. The episode reports vehicle to load arriving for the Model Y through an $80 accessory, with launch edition cars and builds as late as August 2025 excluded, and the required power conversion hardware sold as a retrofit at about $1,700 to $1,800 fitted. For context on how differently rivals are pricing driver assistance, Edmunds notes in its own comparison that FSD now runs $99 a month.
Mild, Medium and Spicy: What Rivian's Three Hands-Free Settings Actually Change
David Vaught took an R2 through city streets, back roads and interstate while switching Universal Hands-Free between its three aggression settings, and his conclusion is that following distance carries almost all of the difference. In town he counts roughly four car lengths in mild, two to two and a half in medium, and about one and a half in spicy. On the highway with a set speed of 85 he estimates ten car lengths in mild against under three in spicy. Lane changes barely move: he could not tell medium from spicy in the city, and even mild was acceptable.
His sharpest criticism is low-speed acceleration in mild, which he found slow enough from a stop that he considers it a hazard in traffic, while the same setting felt fine closing 70 to 80 mph. He also reports the system holds a brake longer than a human would when a car ahead turns out of the lane, an automatic lane change that timed out on a non-highway road, and a spicy lane change it aborted mid-gap.
Worth noting alongside Rivian's autonomy roadmap: none of these settings stop for traffic lights or stop signs yet, which he says he is waiting on.
Four EVs Driven Until They Stopped, and the Recovery Truck That Followed
While Out of Spec ran a constant 70 mph range test to complete depletion, Dave followed with a trailer-mounted DC fast charger and picked the cars up one by one. The R2 stopped first, the video reports, taking about 66 kW at the bottom of its pack, rising toward 68.7 as voltage recovered, and took on 13 kWh in twelve minutes. The Model Y is described as holding 70 mph right up to the moment it quit, with no limp mode, for a run of 317 miles.
The interesting part is what each car does once it is empty. By their account Tesla's state of charge stays inaccurate until the buffer refills, and the team tested whether it would move before reaching 10 percent, finding that it would. The R1S was the problem child: its 12 volt system collapsed, and Kyle explains on camera that the high voltage contactors opened, the DC to DC converter would not restart, and load shedding shut the car down until a forced restart brought systems up in sequence. He describes it as an edge case rather than a daily fault, but says a car should handle its own recovery.
Efficiency numbers surfaced in passing: the R1S covered about 360 miles on 138 kWh drawn, while the Cayenne, given as 108 kWh usable, went 346. Full results are on the Out of Spec Reviews channel.
Edmunds Scored the R2 and the Model Y Dead Even, Then Picked the Rivian
Emme Hall put a fully loaded R2 Launch Edition at about $60,000 against Edmunds' long-term Model Y, bought in March 2025 for just over $60,000 with lifetime full self-driving included. On the Edmunds range loop the R2 returned 304 miles against an EPA figure of 330, while the Model Y matched its 327. She calls the efficiency gap wider than the 20 mile difference suggests, since Edmunds weights efficiency heavily for EVs, and the two cars finished tied on score, both rated excellent and both second overall in the EV SUV rankings.
The R2 wins on packaging: two glove boxes, a digital gauge cluster the Model Y does not have, more rear legroom, a 5.2 cubic foot frunk, a 120 volt outlet rated to 1,900 watts, a rear window that drops, and 4,400 pounds of towing against 3,500. The Model Y is quieter, tighter and better steered, with a cheaper subscription at $10 a month against $15. Neither offers CarPlay or Android Auto.
Timing matters here. The cheaper R2 trims are not out yet, with the roughly $45,000 standard version possibly waiting until spring 2027, and Rivian quotes 345 miles for the coming rear-drive model. Hall would take the R2 home, noting the R1T in their fleet improved through software over a year.
Bottom line: If you want an EX90, wait for a 2026 build and confirm it on the sticker, because the two cars share a badge and very little else, and cheap 2025s are about to look tempting for exactly the wrong reason. Between the Edmunds pair, the R2 is the more interesting car and the Model Y is the more finished one, which is the honest version of a tie. And treat any promised hardware upgrade as worth zero until someone publishes a price for it. A driveway retrofit is a proof, not a plan.
Commentary on third-party videos. Figures and claims are as presented in the sources and have not been independently verified. Spotted an error? Tell us and we will correct it.