Alpamayo 1
Bridging reasoning and action prediction for generalizable autonomous driving
Alpamayo 1 is an advanced AI/ML vision-language-action model designed for autonomous driving research. It combines Chain-of-Causation (CoC) reasoning with trajectory prediction to enable explainable and generalizable autonomous driving in complex scenarios.Following the release of NVIDIA Alpamayo at CES 2026, Alpamayo-R1 has been renamed to Alpamayo 1.
Key features
Chain-of-Causation reasoning
Generates explainable reasoning traces that describe the causal factors influencing driving decisions
Vision-Language-Action architecture
10B parameter model combining Cosmos-Reason backbone with specialized action expert
Long-horizon prediction
Predicts 6.4-second trajectory horizons with 64 waypoints at 10 Hz frequency
Multi-camera processing
Processes multi-camera video inputs with egomotion history for comprehensive scene understanding
What is Chain-of-Causation?
Chain-of-Causation (CoC) is a reasoning approach that generates natural language explanations of the causal factors affecting driving decisions. Unlike traditional black-box models, Alpamayo 1 produces both trajectory predictions and reasoning traces that explain why specific actions were chosen.Model specifications
| Specification | Details |
|---|---|
| Parameters | 10 billion |
| Architecture | Vision-Language-Action with Cosmos-Reason backbone |
| Prediction horizon | 6.4 seconds |
| Waypoints | 64 waypoints at 10 Hz |
| Input | Multi-camera video + egomotion history |
| Output | Trajectory predictions + reasoning traces |
| GPU requirement | NVIDIA GPU with ≥24 GB VRAM |
Use cases
Alpamayo 1 is designed for research and development in autonomous vehicles:- End-to-end driving research: Use as a backbone for autonomous driving experiments
- Reasoning-based auto-labeling: Leverage CoC reasoning for dataset annotation
- Long-tail scenario evaluation: Test model behavior in rare and complex driving situations
- Explainable AI research: Study interpretable decision-making in autonomous systems
Get started
Installation
Set up your environment and install Alpamayo 1
Quickstart
Run your first inference and generate trajectories
Research background
Alpamayo 1 implements the architecture described in the paper “Alpamayo-R1: Bridging Reasoning and Action Prediction for Generalizable Autonomous Driving in the Long Tail”. The model focuses on core supervised learning components, including:- Chain-of-Causation reasoning with hybrid auto-labeling
- Vision-Language-Action architecture combining language understanding with trajectory prediction
- 6.4-second horizon trajectory prediction at 10 Hz
The current release (v1.0) does not include RL post-training, route conditioning, or general VQA capabilities described in the paper. These features may be included in future releases.
License
Alpamayo 1 is released with a dual license structure:- Inference code: Apache License 2.0
- Model weights: Non-commercial license (research and evaluation only)