NVIDIA Releases Alpamayo 2 Super as a Frontier Open Model for Robotaxis and Autonomous Vehicles

NVIDIA Alpamayo 2 Super was released as a commercial open model for the development of robotaxis and autonomous vehicles. Built on NVIDIA Cosmos 3 Super Reasoner and post-trained via reinforcement learning, it supports functions essential for autonomous driving—such as situational understanding, causal reasoning, action selection, and safe path generation—within a single foundation model.

The model is available on Hugging Face under the Linux Foundation's OpenMDW-1.1 permissive license. Developers, automakers, truck manufacturers, and component suppliers can perform fine-tuning, develop derivative models, and commercially redistribute based on their own data and driving policies.

The OpenMDW license also applies to the existing Alpamayo model family, allowing a seamless transition from R&D to commercial deployment without separate permission. Companies can maintain control over their own data and infrastructure while developing specialized models without retraining the foundation model from scratch.

Alpamayo 2 Super is used in cloud development environments to generate high-quality inference traces, synthetic training data, and teacher outputs. Subsequently, a cloud-to-car workflow can be established by performing knowledge distillation into a model family that includes cost-efficient models such as Alpamayo 1.5 and Alpamayo 1, optimizing them for real-time inference in vehicles.

It also demonstrated high performance in benchmarks.

• Ranked 1st among approximately 40 evaluated models on the autonomous driving reasoning benchmark LingoQA.
• Outperformed Qwen2.5-VL 72B by 17.0 points, Gemini 2.5 Pro by 15.1 points, and GPT-4o by 23.2 points in NVIDIA's Lingo-Judge evaluation.
• Ranked 1st in all autonomous driving benchmarks evaluated by NVIDIA.

Alpamayo 2 Super is three times larger than Alpamayo 1.5 and Alpamayo 1, which have 10 billion parameters, aiming to improve the ability to generalize complex reasoning from rare cases.