NVIDIA published a blog post yesterday (August 4), announcing the launch of the open-source Alpamayo 2 Super model for autonomous vehicles. It is now available for commercial use under an open commercial license and offers leading multitask autonomous driving reasoning capabilities.
NVIDIA pointed out in the blog post that the challenges autonomous vehicles face are not everyday driving. The biggest challenges are rare, complex, and difficult-to-predict road conditions.
NVIDIA said that vehicles must not only identify objects and predict motion, but also understand scenes, reason about causal relationships, select the correct actions, and translate decisions into safe and comfortable driving paths. This process must be completed in real time while also being easy for developers to inspect, validate, and trust.
To address these autonomous driving scenarios, NVIDIA has enhanced and launched Alpamayo 2 Super. NVIDIA said the model advances the autonomous driving ecosystem in two ways: open commercial licensing and leading multitask capabilities for autonomous driving.
Regarding the open commercial license, Alpamayo 2 Super is now available on Hugging Face under the OpenMDW-1.1 license. This permissive license from the Linux Foundation is designed for the distribution of open AI models and covers fine-tuning, derivative models, and commercial redistribution.
Autonomous driving developers, automakers, truck manufacturers, and suppliers can adapt Alpamayo based on their own data, driving strategies, and deployment plans.
NVIDIA said this openness allows autonomous driving researchers and companies to retain control over their own data and infrastructure, while also preserving the value created through specialized models and accumulated experience.
NVIDIA said the OpenMDW license now covers the entire Alpamayo model series, allowing developers to use any model for commercial deployment without requiring additional licenses.
In terms of multitask capabilities, Alpamayo 2 Super delivers the highest reasoning and driving performance in the Alpamayo series, making it suitable for multimodal autonomous driving development.
Meanwhile, Alpamayo 1.5 and Alpamayo 1 offer lower-cost options suitable for cloud-based development and model distillation. The distilled models can then be optimized for efficient real-time inference in production vehicles.
In terms of performance, Alpamayo 2 Super ranks No. 1 on the LingoQA autonomous driving reasoning benchmark. In the Lingo-Judge test, Alpamayo 2 Super leads Qwen2.5-VL 72B by 17.0 points and Gemini 2.5 Pro by 15.1 points.
Alpamayo 2 Super is built on **NVIDIA Cosmos 3 Super Reasoner** and trained through reinforcement learning post-training.