NVIDIA Commercialises Its Autonomous Driving AI Model

NVIDIA has commercialised its Alpamayo 2 Super autonomous driving model under an open licence, allowing automakers, robotaxi operators, and suppliers to build, fine-tune, and deploy the model without seeking additional permissions.

The company said that the Alpamayo family has surpassed 500,000 downloads on Hugging Face.

It released Alpamayo 2 Super under the Linux Foundation's OpenMDW-1.1 licence, which permits fine-tuning, derivative models, and commercial redistribution. NVIDIA is also extending the same licence across the entire Alpamayo model family, which had previously been available primarily for research and development.

“Earlier Alpamayo releases were initially introduced for R&D. The OpenMDW license is now being applied across the entire Alpamayo model family so developers can deploy any of the models commercially without requiring additional permissions,” NVIDIA said in a statement.

The company said the move provides autonomous vehicle developers with greater control over their data, infrastructure, and deployment strategies while reducing the cost of developing production-ready systems.

Alpamayo 2 Super is built on NVIDIA Cosmos 3 Super Reasoner and post-trained using reinforcement learning. According to NVIDIA, the model is intended to handle complex driving scenarios by combining perception, reasoning, and planning within a single foundation model.

For each driving scenario, the model generates five outputs: a planned driving trajectory, a chain-of-causation explanation for its decision, a high-level driving action such as yielding or changing lanes, reasoning-based labels for training data, and visual question-answering responses linked to specific regions in camera images.

“Developers can tie what the model observed to the action it selected, making decisions easier to understand, critique, and validate,” the company said.

NVIDIA said the reasoning traces integrate with its Halos safety validation workflows and support AI safety processes aligned with ISO/PAS 8800 requirements.

The company also positioned Alpamayo 2 Super as part of a cloud-to-car development workflow. It said the model can generate reasoning traces, synthetic training data, and teacher outputs in cloud environments, while smaller Alpamayo models can be distilled and optimised for real-time inference inside production vehicles.

“For AV programmes, that means frontier-scale reasoning in the cloud and efficient, specialised models in the vehicle—a more sustainable way to scale safe autonomy into commercial fleets,” NVIDIA said.

In NVIDIA’s benchmark testing, Alpamayo 2 Super ranked 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 points, Gemini 2.5 Pro by 15.1 points and GPT-4o by 23.2 points.

The company said Alpamayo 2 Super has three times the scale of its 10-billion-parameter Alpamayo 1 and Alpamayo 1.5 models, enabling better reasoning across rare driving situations involving multiple road users. The model also processes inputs from front, side, and rear cameras to generate a 360-degree understanding of the driving environment.

Alongside Alpamayo 2 Super, NVIDIA highlighted its broader autonomous driving software stack, including AlpaSim for closed-loop simulation, AlpaGym for reinforcement learning, Physical AI Open Datasets, and open training recipes with an autolabelling pipeline.