Back to feed
News Story
SSignal87
NVIDIA AI Blog
1 sources

NVIDIA Alpamayo 2 Super, the Frontier Open Model for Robotaxis and Autonomous Vehicles, Now Available for Commercial Use

NVIDIA has announced that Alpamayo 2 Super, its open reasoning model for autonomous driving, is now available for commercial use under the Linux Foundation's permissive OpenMDW-1.1 license, which permits fine-tuning, derivative models, and commercial redistribution. Built on NVIDIA Cosmos 3 Super Reasoner and post-trained with reinforcement learning, the model aims to improve handling of rare and complex driving scenarios. This move is expected to expand access to advanced AI for AV developers and support safer, more transparent autonomous vehicle deployment.

SynthePulse Insight · AI deep reading

NVIDIA Alpamayo 2 Super: How Open Licensing and Cutting-Edge Reasoning Are Reshaping Autonomous Driving's Long-Tail Challenge

Version 1 · 1 source

NVIDIA releases Alpamayo 2 Super with open commercial licensing and leading reasoning capabilities, aiming to tackle the most challenging long-tail scenarios in autonomous driving and pave the way for large-scale robotaxi deployment.

  • Alpamayo 2 Super is now commercially available under the Linux Foundation's OpenMDW-1.1 license, permitting fine-tuning, derivative models, and commercial redistribution.
  • Built on NVIDIA Cosmos 3 Super Reasoner and post-trained with reinforcement learning, it ranks first on the LingoQA benchmark.
  • Alpamayo 2 Super has three times the parameters of Alpamayo 1.5 and Alpamayo 1, enhancing reasoning generalization from sparse examples.
  • The model simultaneously outputs five tightly coupled results: trajectories, causal chains, meta-actions, auto-labeling, and visual question answering, improving decision interpretability.
  • The Alpamayo series has surpassed 500,000 downloads on Hugging Face, making it the most popular open-source reasoning model for autonomous driving on the platform.
Open section navigationOpen Commercial Licensing: From R&D to Deployment

Open Commercial Licensing: From R&D to Deployment

NVIDIA Alpamayo 2 Super was announced for commercial use on August 4, 2026, under the Linux Foundation's OpenMDW-1.1 license, which covers fine-tuning, derivative models, and commercial redistribution, allowing developers, automakers, and suppliers to adapt the model to their own data.

Previously, the Alpamayo series was limited to research use; now the entire series adopts the OpenMDW license, enabling developers to deploy commercially without additional licensing, providing a direct path from adaptation to deployment.

Open weights make it economically viable for teams to avoid training base capabilities from scratch and to avoid paying frontier model costs for every task.

Benchmark-Leading Reasoning Capabilities

In NVIDIA's testing, Alpamayo 2 Super ranks first on the LingoQA autonomous driving reasoning benchmark, surpassing 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 has three times the parameters of Alpamayo 1.5 and Alpamayo 1, aiding better reasoning generalization from sparse examples, which is crucial for rare multi-agent interactions.

The model handles full surround-view camera inputs, fusing front, rear, left, and right perspectives to enhance understanding of lane changes, merges, unprotected turns, and complex intersections.

Multi-Task Foundation Model: Interpretable Decision Outputs

Alpamayo 2 Super generates five tightly coupled outputs for each driving scenario: trajectory, causal chain (CoC), meta-actions, reasoning auto-labeling, and visual question answering with 2D visual grounding.

These outputs allow developers to link what the model observes to the actions it selects, making decisions easier to understand, critique, and validate.

CoC trajectories integrate with NVIDIA Halos safety validation workflows and support AI safety compliant with ISO/PAS 8800, providing a foundation for AV safety engineering.

Cloud-to-Car Deployment Workflow

Alpamayo 2 Super supports cloud-based development workflows, generating high-quality reasoning trajectories, synthetic training data, and teacher outputs for model distillation.

Distilled models can be optimized for real-time inference in production vehicles, forming a complete cloud-to-car workflow that balances frontier reasoning and scalable deployment.

Alpamayo 1.5 and Alpamayo 1 offer more cost-effective options for cloud development and model distillation.

Open Ecosystem and Adoption

Alpamayo 2 Super is part of NVIDIA's open autonomous driving ecosystem, including AlpaSim closed-loop simulation, AlpaGym reinforcement learning, Physical AI open datasets, and open training recipes and auto-labeling pipelines.

The Alpamayo series has surpassed 500,000 downloads on Hugging Face, making it the most popular open-source reasoning model series for autonomous driving on the platform.

The model can serve as an auto-labeler, generating CoC labels and visual question answering on proprietary fleet data, compressing annotation cycles from months to days.

Credibility boundary

This article's information primarily comes from NVIDIA's official blog, which is a vendor release. Benchmark results and performance data are NVIDIA's own testing and have not been independently verified.

Insight takeaway

Alpamayo 2 Super offers an interpretable solution for autonomous driving long-tail scenarios through open licensing and frontier reasoning capabilities, but its real-world effectiveness still requires independent verification.

Sources for this version

  1. NVIDIA Alpamayo 2 Super, the Frontier Open Model for Robotaxis and Autonomous Vehicles, Now Available for Commercial UseNVIDIA AI Blog

Primary report

NVIDIA AI Blog

Primary source