NVIDIA released Alpamayo 2 Super, a 34-billion-parameter reasoning model for autonomous driving, for commercial use on August 4, 20267,4. The model weights are available on Hugging Face under OpenMDW-1.1, the Linux Foundation's permissive license covering fine-tuning, derivative models, and commercial redistribution.
The release clears a licensing barrier that had kept the Alpamayo family restricted to research use. NVIDIA says the OpenMDW license now applies across the entire Alpamayo lineup, allowing developers, automakers, truckmakers, and suppliers to deploy any of the models commercially without seeking additional permission. The company describes Alpamayo as the most-adopted open reasoning model family for autonomous driving on Hugging Face, with downloads passing 500,000.
Alpamayo 2 Super is built on NVIDIA's Cosmos 3 Super Reasoner and post-trained with reinforcement learning. The architecture pairs a 32-billion-parameter vision-language backbone with a 2.3-billion-parameter diffusion-based action decoder. The model ingests multi-camera video, text, and the vehicle's motion history, and produces five coupled outputs for each driving situation: a trajectory, a chain-of-causation trace, a meta-action, reasoning auto-labels, and visual question-answering responses tied to specific regions of the camera image through 2D grounding.
NVIDIA positions Alpamayo 2 Super as a teacher model for cloud-based workflows rather than software that would steer a production vehicle. The model handles the heaviest reasoning work in the cloud, generating reasoning traces, synthetic training data, and teacher outputs used to distill smaller models that can run on in-vehicle compute. NVIDIA says the model can compress annotation cycles from months to days when used as an autolabeler on proprietary fleet data.
The model targets long-tail events — rare, multi-agent traffic situations where conventional systems built on object detection and motion prediction tend to struggle5. For each scenario, Alpamayo 2 Super generates a planned trajectory alongside an explanation of the factors behind its decision and a high-level action such as yielding, stopping, or changing lanes. NVIDIA says the reasoning traces integrate with its Halos safety-validation workflows and support engineering aligned with ISO/PAS 8800 requirements.
In NVIDIA's own testing, Alpamayo 2 Super scored 79.2 on the Lingo-Judge metric of LingoQA, which NVIDIA says ranks first among nearly 40 models evaluated. NVIDIA reported a 23.2-point lead over GPT-4o, a 15.1-point lead over Gemini 2.5 Pro, and a 17.0-point lead over Qwen2.5-VL 72B on the same metric. The model card reports an open-loop trajectory error of 0.911 meters at a 6.4-second horizon on 937 challenging samples, and closed-loop evaluation in NVIDIA's AlpaSim simulator across 910 scenarios from its PhysicalAI-AV-NuRec dataset.
NVIDIA first unveiled Alpamayo 1 at CES as what it called the industry's first chain-of-thought reasoning vision-language-action model for AV research. Alpamayo 2 Super runs at roughly three times the parameter count of the 10-billion-parameter Alpamayo 1 and 1.5 models. NVIDIA introduced Alpamayo 2 Super at GTC Taipei on May 31, 2026, with commercial availability promised for the summer.
Grayson Brulte and Rob Grant, co-founders of research firm Autonomy AI, argued on their Road to Autonomy podcast that the open-source release is a platform capture strategy: NVIDIA gives away the model to make it economically irrational for AV developers to build their own foundation model, while locking the downstream pipeline into NVIDIA's Drive Thor silicon and DGX cloud infrastructure2.
ANALYSIS The teacher-model architecture means Alpamayo 2 Super's commercial value flows through NVIDIA's hardware stack at two points — cloud-side training and distillation on DGX, and edge inference on Drive Thor — rather than through direct software licensing revenue.