Alpamayo 2 Super - NVIDIA's Open-Source Autonomous Driving AI Reasoning Model

What is Alpamayo 2 Super

Alpamayo 2 Super is NVIDIA's open-source autonomous driving AI reasoning model. The model is built on NVIDIA Cosmos 3 Super Reasoner and features 360° environmental perception, advanced driving decision-making, and automatic reasoning label generation capabilities, ranking first on the LingoQA autonomous driving reasoning benchmark. The model is released on Hugging Face under the permissive OpenMDW-1.1 license, supporting fine-tuning, derivatives, and commercial redistribution, while also offering Alpamayo 1.5 / 1 versions for cloud development and vehicle-side distillation deployment, targeting the Robotaxi and autonomous vehicle domains.

Alpamayo 2 Super's Main Features

• 360° Environmental Perception: Achieves high-precision recognition and understanding of omnidirectional scenes around the vehicle, covering complex real-world road conditions.
• Advanced Driving Decision Reasoning: Performs causal reasoning based on scene context to output safe and comfortable driving paths and action decisions.
• Automatic Reasoning Label Generation: Can automatically generate driving scene reasoning labels, helping developers quickly build training datasets.
• Multimodal Autonomous Driving Development: Supports fusion of multimodal inputs such as vision and text, unifying perception and decision-making tasks.
• Model Distillation Support: Provides Alpamayo 1.5 / 1 versions, supporting cloud development and distillation into efficient inference models for mass-production vehicles.

Alpamayo 2 Super's Technical Principles

• Base Architecture: Built on the NVIDIA Cosmos 3 Super Reasoner super reasoning model, inheriting its multimodal understanding and long-chain reasoning capabilities.
• Reinforcement Learning Post-Training: Performs autonomous driving-specific optimization of the model through RL Post-Training, enhancing causal reasoning, risk anticipation, and path planning capabilities.
• End-to-End Multi-Task Architecture: Adopts a unified architecture to simultaneously handle core autonomous driving modules such as perception, prediction, and planning, achieving end-to-end reasoning from scene understanding to driving decisions.
• Data-Driven Iteration: Conducts iterative training based on large-scale real driving data and simulation environments, covering rare and complex road conditions to improve generalization capability.
• Model Distillation System: Supports distillation from Alpamayo 2 Super to the 1.5 / 1 versions, compressing model size while maintaining performance, adapting to low-compute real-time inference needs on vehicles.

Follow WeChat and reply "open source" to join theAI Open Source Project Exchange Group

How to Use Alpamayo 2 Super

• Obtain Model Weights: Visit the official Hugging Face repository to download the model weights and configuration files of Alpamayo 2 Super.
• Confirm Commercial License: Read and comply with the terms of the OpenMDW-1.1 open-source license to ensure compliance for fine-tuning, derivatives, and commercial deployment.
• Prepare Driving Data: Integrate your own vehicle sensor data, driving strategies, and scene annotations, and complete the training and validation environment setup.
• Model Fine-Tuning and Distillation: Perform domain fine-tuning of Alpamayo 2 Super based on your own data, or distill it to Alpamayo 1.5 / 1 to adapt to vehicle-side compute.
• Integration and Deployment Validation: Integrate the optimized model into the vehicle perception-decision-control pipeline, and validate system stability through simulation testing and real-vehicle road tests.

Core Advantages of Alpamayo 2 Super

• Open Source and Commercial Use: Adopts the Linux Foundation OpenMDW-1.1 permissive license, supporting fine-tuning, derivatives, and commercial redistribution without additional authorization.
• Leading Inference Performance: Ranked first on the LingoQA autonomous driving reasoning benchmark, significantly outperforming Qwen2.5-VL 72B and Gemini 2.5 Pro in the Lingo-Judge test.
• Full-Stack Coverage: Provides a complete model family from cloud high-compute inference (Alpamayo 2 Super) to in-vehicle lightweight deployment (Alpamayo 1.5 / 1).
• Data Sovereignty Preserved: Developers have full control over their own data and infrastructure, and the model's openness safeguards the accumulation of enterprise proprietary knowledge.

Alpamayo 2 Super Project Address

• HuggingFace Model Repository：https://huggingface.co/nvidia/Alpamayo2-Super