NVIDIA Open-Sources Alpamayo 2 Super to Power Next-Generation Autonomous Vehicles

NVIDIA has released Alpamayo 2 Super for commercial use, expanding its open autonomous driving AI portfolio with a new reasoning model designed for robotaxis and other autonomous vehicles. Available today under the Linux Foundation's OpenMDW-1.1 license, the model gives automakers, suppliers and developers permission to fine-tune, modify and commercially deploy the technology without requiring additional licensing approvals.

Alpamayo 2 Super is built on NVIDIA Cosmos 3 Super Reasoner and post-trained with reinforcement learning. The company says the model is designed to address complex driving situations by combining multiple autonomous driving capabilities within a single foundation model while supporting a cloud-to-car development workflow.

The commercial licensing marks a significant expansion for the Alpamayo family. Earlier releases were introduced primarily for research and development, but NVIDIA is now applying the OpenMDW-1.1 license across the entire model family, allowing developers to adapt the models using their own fleet data, driving policies and deployment strategies.

According to NVIDIA, the approach enables developers to retain control over proprietary data and infrastructure while reducing the need to retrain foundation models from scratch. The company says Alpamayo 2 Super is intended for frontier-scale reasoning in cloud environments, while Alpamayo 1.5 and Alpamayo 1 remain lower-cost options for cloud development and model distillation before deployment in production vehicles.

NVIDIA says Alpamayo 2 Super ranks first on the LingoQA autonomous driving reasoning benchmark among nearly 40 evaluated models. Using the company's Lingo-Judge metric, it outperformed Qwen2.5-VL 72B by 17.0 points, Gemini 2.5 Pro by 15.1 points and GPT-4o by 23.2 points. The company also says the model ranked first across all autonomous driving benchmarks it evaluated.

The new model is approximately three times larger than the 10-billion-parameter Alpamayo 1 and Alpamayo 1.5 models. NVIDIA says the increased scale improves the model's ability to generalize from limited examples and reason through uncommon driving situations involving multiple agents. The model processes 360-degree camera inputs by combining views from the front, sides and rear of a vehicle to better understand scenarios such as lane changes, merges, unprotected turns and complex intersections.

For each driving scenario, Alpamayo 2 Super generates five connected outputs: a planned driving trajectory, a chain-of-causation explanation describing its reasoning, a meta-action such as yielding or changing lanes, reasoning auto-labels for training and validation, and visual question-answering responses with 2D visual grounding tied to specific regions in camera images.

NVIDIA says these outputs are designed to make model decisions easier to inspect and validate throughout autonomous vehicle development. The company also says chain-of-causation traces integrate with its Halos safety validation workflows and support AI safety aligned with ISO/PAS 8800 requirements.

The model can also function as an autolabeling system for proprietary fleet data. NVIDIA says it can generate chain-of-causation labels and visual question-answering annotations with image grounding, allowing developers to transform raw driving footage into training data while reducing annotation cycles from months to days.

Beyond planning and data labeling, Alpamayo 2 Super supports scene understanding, model critiquing and knowledge distillation, enabling developers to use a single foundation model across multiple stages of autonomous vehicle development.

The release is part of NVIDIA's broader open autonomous driving ecosystem, which also includes AlpaSim for closed-loop simulation, AlpaGym for reinforcement learning, NVIDIA Physical AI Open Datasets, open training recipes and an autolabeling pipeline. NVIDIA says the Alpamayo family has surpassed 500,000 downloads on Hugging Face, making it the platform's most-adopted open reasoning model family for autonomous driving.

See more here:

This analysis is based on reporting from NVIDIA.

Image courtesy of NVIDIA.

This article was generated with AI assistance and reviewed for accuracy and quality.