NVIDIA releases Alpamayo 2 Super, a 30 billion parameter open reasoning model for autonomous driving with state-of-the-a

NVIDIA Alpamayo 2 Super is now available for commercial use as part of the Alpamayo family, the most adopted open reasoning models for autonomous driving on Hugging Face. The model is designed to handle the long-tail events and rare, complex driving scenarios that are the hardest problems for autonomous vehicles, requiring the system to understand situations, reason about cause and effect, choose the right action and execute safe, comfortable paths in real time.

Built on NVIDIA Cosmos 3 Super Reasoner and post trained with reinforcement learning, Alpamayo 2 Super advances the autonomous vehicle ecosystem through open commercial licensing and leading multitask capabilities. The model is available on Hugging Face under OpenMDW 1.1, the Linux Foundation's permissive license for open AI model distributions, which covers fine tuning, derivative models and commercial redistribution. This allows autonomous vehicle developers, automakers, truckmakers and suppliers to adapt Alpamayo to their own data, driving policies and deployment strategies while keeping control of their proprietary data, infrastructure and the specialized models they create.

Alpamayo 2 Super offers 3x the scale of the 10 billion parameter Alpamayo 1.5 and Alpamayo 1 models, with 30 billion parameters that help the model better generalize reasoning from sparse examples, a critical capability for rare multi agent interactions. The model reasons over full surround camera coverage from front, sides and rear, enabling 360 degree context for understanding lane changes, merges, unprotected turns and complex intersections where risks commonly arise.

For each driving situation, Alpamayo 2 Super produces five tightly coupled outputs: a trajectory describing the vehicle's planned path, a chain of causation trace that explains the reasoning behind the decision, a meta action such as yield or lane changes that captures the model's intent, reasoning auto labels that generate annotations for training and validation data, and visual question answering responses with 2D visual grounding that link the model's answers to specific regions in camera images. These outputs offer developers insight into the model's decision making process, making decisions easier to understand, critique and validate. Chain of causation traces integrate with NVIDIA Halos safety validation workflows and support AI safety aligned with ISO/PAS 8800 requirements.

In NVIDIA testing using the Lingo Judge metric, Alpamayo 2 Super ranks first on LingoQA, an autonomous driving reasoning benchmark among nearly 40 models evaluated, outperforming Qwen2.5 VL 72B by 17.0 points, Gemini 2.5 Pro by 15.1 points and GPT 4o by 23.2 points. The model also ranks first across all autonomous driving benchmarks evaluated by NVIDIA.

Beyond planning and auto labeling, Alpamayo 2 Super supports scene understanding, model critiquing and knowledge distillation as multitask capabilities, enabling developers to use a single foundation model across more of the development stack. The model can be deployed as an autolabeler to generate chain of causation labels and perform visual question answering on proprietary fleet data, transforming raw driving clips into richer training data and compressing annotation cycles from months to days. This cloud to car workflow combines frontier scale reasoning in cloud based development workflows where developers can generate high quality reasoning traces and synthetic training data with scalable, efficient deployment across commercial autonomous vehicle fleets.

Alpamayo 2 Super is part of a broader family of NVIDIA open models, frameworks and datasets for autonomous driving development, including AlpaSim for closed loop simulation, AlpaGym for high throughput reinforcement learning and Physical AI Open Datasets for training and testing, along with open training recipes and an autolabeling pipeline. The Alpamayo family has surpassed 500,000 downloads on Hugging Face, reinforcing its position as the most adopted open reasoning model family for autonomous driving on the platform.