이미지: METAL LAB 생성
Summary
- NVIDIA has released Alpamayo 2 Super, an open reasoning model for autonomous driving, under a license that permits commercial use
- The company said the model scored ahead of Qwen2.5-VL 72B, Gemini 2.5 Pro, and GPT-4o on the LingoQA benchmark
- Through the Alpamayo 2 Super, 1.5, and 1 family, NVIDIA presented a workflow spanning from cloud-based reasoning to lightweight in-vehicle deployment
- 모델명
- NVIDIA Alpamayo 2 Super
- 기반 모델
- NVIDIA Cosmos 3 Super Reasoner, 강화학습 후처리
- 라이선스
- Hugging Face 공개, OpenMDW-1.1 (Linux Foundation 허가 라이선스)
- 라이선스 허용 범위
- 파인튜닝, 파생 모델, 상업적 재배포
- 벤치마크
- LingoQA 자율주행 추론 벤치마크 약 40개 모델 중 1위
- 성능 비교(Lingo-Judge 기준)
- GPT-4o 대비 +23.2점, Qwen2.5-VL 72B 대비 +17.0점, Gemini 2.5 Pro 대비 +15.1점
- 모델 패밀리
- Alpamayo 2 Super, Alpamayo 1.5, Alpamayo 1
A Fully Open Model for Commercial Use
NVIDIA has released Alpamayo 2 Super, an open reasoning model for robotaxis and autonomous vehicles (AVs), in a form that permits commercial use. The model belongs to the Alpamayo family, the most widely adopted family of open reasoning models for autonomous driving on Hugging Face. It was built through reinforcement learning post-training on top of NVIDIA Cosmos 3 Super Reasoner.
The model is distributed on Hugging Face under OpenMDW-1.1, a permissive license from the Linux Foundation. The license covers fine-tuning, creation of derivative models, and commercial redistribution, allowing automakers, truck manufacturers, and parts suppliers to adapt the model to their own data and driving policies. Unlike earlier Alpamayo versions, which were first released for research and development purposes, NVIDIA said the OpenMDW license now applies across the entire family, opening a path to commercial deployment without requiring separate approval.
Reasoning in the Cloud, Lightweight in the Vehicle
NVIDIA positioned Alpamayo 2 Super as a model for performing frontier-level reasoning within cloud-based development workflows. Developers can use the model to generate high-quality reasoning traces, synthetic training data, and teacher-model outputs for distillation. The resulting lightweight models are then optimized for real-time inference inside actual vehicles.
Within the family, Alpamayo 2 Super offers the highest reasoning and driving performance, while Alpamayo 1.5 and Alpamayo 1 are presented as more cost-efficient options for cloud-based development and model distillation. NVIDIA described this as a "cloud-to-car" structure: "frontier-level reasoning in the cloud, efficient and specialized models in the vehicle."
Ranked First on LingoQA Benchmark
NVIDIA said that in its own testing, Alpamayo 2 Super ranked first among roughly 40 models on LingoQA, a benchmark for autonomous driving reasoning. Based on the Lingo-Judge metric, the company reported that it scored 17.0 points higher than Qwen2.5-VL 72B, 15.1 points higher than Gemini 2.5 Pro, and 23.2 points higher than GPT-4o.
| Comparison Model | Score Gap vs. Alpamayo 2 Super |
|---|---|
| GPT-4o | +23.2 pts bar:100 |
| Qwen2.5-VL 72B | +17.0 pts bar:73 |
| Gemini 2.5 Pro | +15.1 pts bar:65 |
These figures are based on NVIDIA's own internal testing, and no third-party verification has been confirmed separately.
Related news on autonomous driving models has also been covered in metallab.ai's article on open-source AI models.
Openness Reshaping the Ecosystem
NVIDIA said Alpamayo 2 Super is part of a broader strategy to expand open models, datasets, and tools for autonomous driving. By releasing open weights, development teams can choose models suited to their needs without having to retrain all underlying capabilities from scratch or pay frontier-model-level costs for every task. NVIDIA called this "a more sustainable way to scale safe autonomous driving across commercial AV fleets."
