NVIDIA Alpamayo 2 Super Open Model for Robotaxis and AVs Now Commercially Available

NVIDIA has released Alpamayo 2 Super, an open reasoning model for robotaxis and autonomous vehicles (AVs), now available for commercial use. Built on NVIDIA Cosmos 3 Super Reasoner and post-trained with reinforcement learning, it is part of the Alpamayo family, the most-adopted open reasoning models for autonomous driving on Hugging Face. The model is designed for rare, complex driving situations, combining situation understanding, cause-and-effect reasoning, and safe path planning in real time.

Alpamayo 2 Super is distributed under OpenMDW-1.1, a permissive license from the Linux Foundation that covers fine-tuning, derivative models, and commercial redistribution. AV developers, automakers, truckmakers, and suppliers can adapt it to their own data, driving policies, and deployment strategies while keeping control of proprietary fleets and infrastructure. Earlier Alpamayo releases were limited to research; the new license applies across the entire family, enabling a direct path from adaptation to production. Open weights allow teams to build on advanced reasoning without re-training every capability or paying frontier-model costs. The family supports cloud-based development, where Alpamayo 2 Super generates reasoning traces and teacher outputs for distillation, yielding models that can be optimized for real-time inference in production vehicles — a cloud-to-car workflow; Alpamayo 2 Super is the highest-performing option, while 1.5 and 1 offer cost-efficient alternatives.

Alpamayo 2 Super also ranks first across all autonomous driving benchmarks evaluated by NVIDIA. It offers 3x the scale of the 10-billion-parameter Alpamayo 1.5 and 1 models, improving generalization from sparse examples. The model reasons over full-surround camera views — front, sides, and rear — for richer understanding of lane changes, merges, unprotected turns, and complex intersections. Its chain-of-causation traces integrate with NVIDIA Halos safety-validation workflows and support ISO/PAS 8800 safety engineering. As an autolabeler, it compresses annotation cycles from months to days and also supports scene understanding, model critiquing, and knowledge distillation.