NVIDIA releases Alpamayo 2 Super for commercial autonomous driving

The hardest problem in autonomous driving is not routine scenarios but rare edge cases that are difficult to anticipate and train for. Handling them requires situational awareness, causal reasoning, selecting the correct action and converting it into a safe, comfortable trajectory in real time, while remaining testable and verifiable by developers. NVIDIA is now releasing Alpamayo 2 Super for commercial use as part of the most widely adopted open-model family for autonomous driving on Hugging Face. The model builds on NVIDIA Cosmos 3 Super Reasoner and has been further trained with reinforcement learning.

Alpamayo 2 Super is offered on Hugging Face under OpenMDW-1.1, a permissive license from the Linux Foundation for distributing AI models. The license covers fine-tuning, derivative models and commercial redistribution, allowing car makers, truck manufacturers and suppliers to adapt Alpamayo to their data, driving policies and deployment strategies. This approach lets companies retain control of their data and infrastructure while extracting value through specialized models. Earlier Alpamayo releases were limited to research and development; now OpenMDW applies to the entire model family, enabling commercial deployment of any model without additional permissions.

Open weights make the path economical. Teams can build on advanced reasoning without training basic capabilities from scratch or paying frontier-scale reasoning costs for every task. Alpamayo 2 Super provides frontier-scale reasoning in cloud-based development, where engineers generate high-quality reasoning traces, synthetic training data and teacher outputs for model distillation. Within the family, Alpamayo 2 Super delivers the highest reasoning and driving performance for multimodal autonomous-driving development, while Alpamayo 1.5 and Alpamayo 1 offer more cost-effective options for cloud development and model distillation.

In NVIDIA’s internal benchmarks using the Lingo-Judge metric on LingoQA, a driving-focused reasoning benchmark, Alpamayo 2 Super ranked first among almost 40 evaluated models. 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. These results come from NVIDIA’s own testing, not an independent assessment, and cover driving-specific scenarios. Alpamayo 2 Super also topped every autonomous-driving benchmark that NVIDIA compiled.

The distilled models generated in the cloud are tuned for real-time inference in production vehicles. The entire family supports a cloud-to-vehicle workflow that combines frontier-scale reasoning with scalable deployment to commercial fleets. For autonomous-driving programs, this means cloud-level reasoning paired with efficient, specialized in-vehicle models, a sustainability narrative NVIDIA presents for safely extending autonomy to commercial fleets.