Tesla-leiding
Tesla's own VP said it. HW3 runs a 'smaller model'.
Tesla's VP of Autopilot publicly admitted on X that Hardware 3 vehicles run a 'smaller model' than vehicles with the newer AI4 hardware. Tesla's top FSD engineer confirmed product differentiation himself.
-
Date
-
21 August 2024
We were able to achieve similar performance as the AI4 12.5 release with a relatively smaller model for AI3.
Who Ashok Elluswamy is
Ashok Elluswamy is Tesla's VP of Autopilot. He's the senior engineer responsible for the Full Self-Driving programme, a direct report to Elon Musk, and speaks in that capacity for Tesla on technical FSD matters. When he makes a statement about how FSD works on which hardware, it's a party statement in the legal sense.
What he said
On 21 August 2024, eight months before Tesla EMEA NL Contact rejected the first summation letter from the initiator of this site, Elluswamy wrote on X in response to a discussion about FSD v12.5 and hardware generations:
"We were able to achieve similar performance as the AI4 12.5 release with a relatively smaller model for AI3."
And, following up, about what it took to get similar results at all:
"Deploying the larger model requires us to implement a few kernels in the compiler to emulate the same operations that are natively supported on the AI4 hardware."
What that means (and what it doesn't)
What it says:
- Tesla developed a smaller model of FSD for HW3 (Elluswamy calls it 'AI3', it's the same computer). Not an equivalent model. Not an identical version. A 'relatively smaller' version.
- The performance is described as 'similar'. Not 'equivalent'. Not 'identical'. NotaTeslaApp, in accompanying commentary, added that it's 'not exactly the same'.
- To run the larger (full) model on HW3, Tesla has to implement separate workarounds ('kernels') in the compiler that emulate operations which AI4 supports 'natively'.
What it doesn't say:
- Not that HW3 vehicles receive the full FSD model
- Not that performance is identical to AI4
- Not that there's no difference in safety or functionality
- Not that HW3 can do everything AI4 can
This is the moment Tesla lost deniability
Until August 2024, Tesla could still claim that 'FSD would soon be available for everyone via OTA, on every hardware generation.' With this public statement from its own VP, that changed. From that date on, Tesla publicly knows, and anyone who looks it up publicly knows, that HW3 and AI4 run fundamentally different products.
Tesla Support rejected Mischa Sigtermans on 24 April and 9 May 2025 with the claim that FSD 'depends on regulation and software development'. That rejection directly contradicts what Tesla's own FSD lead had said publicly eight months earlier. That makes Tesla's response to the summations not just unfortunate. It makes it legally contestable under article 6:193c of the Dutch Civil Code (misleading commercial practice): Tesla issued a standard statement which it knew at the time of giving wasn't true.
Verification
The X post was reported by the independent Tesla tracker NotaTeslaApp, which published a detailed technical breakdown. See:
https://www.notateslaapp.com/news/2204/tesla-hw3-runs-smaller-fsd-model-than-ai4-performance-to-be-similar
For legal proceedings it's advisable to archive the original tweet itself (via archive.today or notarial confirmation), so Tesla can't later claim the post was edited or deleted.