NVIDIA unveils Alpamayo 2 Super, a 32‑billion‑parameter vision language action model designed to enable Level 4 autonomous driving in robotaxi scenarios, with advanced environmental understanding and safe trajectory decisions.

The model acts as a teacher model capable of distilling knowledge into smaller, optimized in-vehicle models suitable for hardware like DRIVE AGX Thor.

Alpamayo 2 Super marks a major scale-up from Alpamayo 1, which had 10 billion parameters and debuted at CES 2026, expanding from open‑weights trajectory generation to full reasoning and planning.

The release includes forward‑looking statements and risk disclosures, noting that availability may vary by region and market conditions.

The company indicates the announcement occurred on June 1, 2026, with availability expected in summer 2026 on GitHub and Hugging Face.

Uber is highlighted as having shifted from direct vehicle technology development to playing a platform role since 2020.

An open‑source CoC Auto-Labeling Pipeline will be released to generate causally linked training labels without human annotation, supporting scalable embodied reasoning model training.

The Alpamayo ecosystem now supports a continuous path from open‑loop pretraining to closed‑loop refinement, integrating the CoC Auto‑Labeling Pipeline for causal, decision-grounded labels.

Open‑source deployment is buoyed by strong user interest, with Alpamayo’s prior versions amassing over 400,000 downloads.

The broader industry trend is toward end‑to‑end autonomous driving reasoning, aiming to improve handling of unpredictable road scenarios beyond simple trajectory generation.

There is no crypto or blockchain integration mentioned; Nvidia emphasizes traditional tech and automotive applications.

NVIDIA underscores safety validation, regulatory collaboration, and scaling Level 4 AV deployment across millions of vehicles as core benefits of the Alpamayo ecosystem.