FSD Beta V12.1 leak reveals shift toward end-to-end neural network

An internal FSD update rollout for Tesla employees has confirmed that the company is testing FSD Beta V12.1, which shifts toward the end-to-end neural network.

The latest changes appeared a few days ago via update 2023.44.30.10, it’s an incremental push over the 2023 Holiday update with version 2023.40.30 released last week.

Surprisingly, the update changelog mentions FSD Beta v12.1 with the following change.

The changelog says that the FSD Beta v12 upgrades the city-streets driving stack to a single end-to-end neural network trained on millions of video clips, replacing over 300k lines of explicit C++ code.

Tesla FSD V12.1

Changelog is quite small but Teslacope suggests that this rollout is limited some Tesla employees in the US.

Meanwhile, the changelog itself can’t be ignored because it has a major change toward “unsupervised” FSD.

Since the very beginning, Tesla FSD, aka Full Self Driving, has been collecting driving data to train its neural network. However, the company so far has relied on supervised training using a bunch of manual instructions via programmers.

With that being said, the FSD Beta V12 promises to take a new route into the world of autopilot. Last month, Tesla CEO, Elon Musk said that FSD version 12 with end-to-end neural nets will “far exceed human safety” even in an unsupervised state.

Despite these statements, the release of FSD Beta V12.1 remains internal and we’ve not yet seen the power end-to-end neural network in action.

Still Beta?

Tesla’s Musk mentioned in June that Version 12 won’t be beta” However, this v12.1 is labeled as “Beta” in the changelog.

It is possible that the past issues with the autopilot recall of 2 million vehicles may have changed Tesla’s strategy around this feature and led the company to keep FSD in test mode until it fully verifies the stability.

For now, there’s no timeline available for the release of FSD Beta V12 for FSD users.