Nvidia launches Alpamayo 2 Super, its most powerful open AI model for robotaxis

Nvidia just dropped what it’s calling the most powerful open reasoning model for autonomous vehicles. Alpamayo 2 Super, unveiled at GTC Taipei on May 31, packs 32 billion parameters and is purpose-built for companies trying to crack level 4 robotaxi systems.

What Alpamayo 2 Super actually does

The model belongs to a class called VLA, or vision-language-action models. Think of it as a system that doesn’t just see the road and spit out a driving path. It reasons about what it’s seeing, plans a response, and then acts on that plan. Previous autonomous driving models have largely been limited to trajectory generation, which is a fancy way of saying they could plot a route but struggled with the weird, unpredictable stuff that happens on real roads.

Alpamayo 2 Super tackles those so-called long-tail scenarios through chain-of-thought reasoning across the full autonomous driving stack.

The jump in scale is notable. Nvidia’s original Alpamayo 1 model, introduced at CES on January 5, 2026, carried 10 billion parameters. Moving to 32 billion in roughly five months represents more than a threefold increase in model size.

Alpamayo 1 shipped with open weights, inferencing scripts, and tooling for video-input trajectory generation. Alpamayo 2 Super extends that foundation into full reasoning and planning territory.

The partner list tells the story

Nvidia highlighted several organizations already working with the Alpamayo model family: Lucid, Jaguar Land Rover, Uber, and Berkeley DeepDrive.

Uber’s involvement is particularly interesting. The company famously sold its own self-driving unit, Advanced Technologies Group, back in 2020 after years of expensive development and a fatal crash in Arizona. Since then, Uber has positioned itself as the platform layer for autonomous vehicles rather than building the technology itself.

One thing conspicuously absent from any of Nvidia’s Alpamayo announcements: crypto or blockchain integration. Nvidia is keeping this initiative firmly in the traditional technology and automotive lane.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.