Alpamayo 2 Super
#8 in AI Inference Hardwarenvidia · v2 · super · since 2026-06-01 · 4× · last seen Jun 29, 2026
36
Momentum
NVIDIA Alpamayo 2 Super is an open-source Vision-Language-Action (VLA) model with 32 billion parameters, built on the NVIDIA Cosmos 3 Super Reasoner architecture and post-trained via reinforcement learning. It is purpose-built for Level 4 autonomous vehicle (robotaxi) development and serves as a teacher model that can be distilled into compact on-vehicle models. The model processes multi-camera video input (360-degree surround view) and outputs driving trajectories, Chain-of-Causation reasoning traces, and Meta-Action predictions for high-level driving maneuvers. Inference code and model weights are planned for release in summer 2026 via GitHub and Hugging Face.
Momentum trend
04.04.03.07.
Features
| Context Window/Max Input | Multi-camera video input (up to 16 cameras per Alpamayo 1 latency study); 360-degree surround view (front, side, rear); processes multi-sensor driving data incl. navigation inputs and driving context |
| Price Tier | Open source / free (non-commercial, no license required; commercial licensing available on request); model weights via Hugging Face, inference code via GitHub |
| Target Platform Type | Cloud/datacenter as teacher model (training & inference); in-vehicle deployment via NVIDIA DRIVE AGX Thor (NVIDIA DRIVE Hyperion platform) after distillation into compact models |