NVIDIA releases 32B Alpamayo 2 Super reasoning model
NVIDIA has unveiled Alpamayo 2 Super, its newest open foundation reasoning model designed specifically for autonomous driving and physical AI applications. Built on the NVIDIA Cosmos 3 backbone with 32 billion parameters—triple the size of its predecessors—Alpamayo 2 Super provides 360-degree vision-language perception across multi-camera setups to output spatial meta-actions and predict vehicle trajectories.
Alpamayo 2 Super demonstrates the rapid shift of autonomous driving architectures toward compute-intensive, end-to-end reasoning foundation models.
- –Massive Parameter Scale: At 32B parameters, Alpamayo 2 Super scales up reasoning capabilities to handle complex, multi-camera driving environments.
- –Teacher Model Distillation: Functions as an offline reasoning engine for auto-labeling and training smaller, real-time edge models deployed in vehicles.
- –Multi-Camera Perception: Natively integrates 360-degree surround-view inputs to output future trajectories and high-level driving decisions.
- –Infrastructure Demand: The exponential rise in compute requirements for training and running complex physical AI models reinforces strong demand for AI cloud providers like Nebius.
DISCOVERED
1d ago
2026-08-04
PUBLISHED
1d ago
2026-08-04
RELEVANCE
AUTHOR
jiwanpark1014