Alpamayo is NVIDIA's 10B parameter vision-language-action model that enables autonomous vehicles to reason step-by-step (via Chain-of-Causation) before predicting safe driving trajectories, making decisions more interpretable and robust in complex scenarios.

Is Alpamayo completely open-source?

Model weights are available for non-commercial research use on Hugging Face, while the inference code is fully open under Apache 2.0 license.

When was Alpamayo released?

The model was initially uploaded to Hugging Face on December 3, 2025, with major public announcement and renaming to Alpamayo 1 during CES 2026 in early January.

What hardware do I need to run Alpamayo locally?

You need at least one NVIDIA GPU with 24GB+ VRAM (such as RTX 4090 or H100) and Linux OS, along with PyTorch, Transformers, and DeepSpeed for optimal inference performance.

Alpamayo

Reasoning Vision Language Action Model for Autonomous Driving

From NVIDIA

Founder: NVIDIA Research Team

Video & Animation

Dec 2025

10K+

Free

Pricing Model

Free

Starting Price

$0/Month

About This AI

Alpamayo (formerly known as Alpamayo-R1-10B) is NVIDIA’s open weights 10B parameter vision-language-action (VLA) model that integrates Chain-of-Causation reasoning with precise trajectory planning.

It processes multi-camera images, text commands, and vehicle motion history to output interpretable reasoning traces explaining decisions and generates future 6.4-second driving trajectories (64 waypoints at 10Hz) for handling complex, long-tail autonomous driving scenarios like rare events, adverse conditions, and intricate interactions.

AI Agents Image To Video Video Editor Video Generator

Pricing

Pricing Model

Free

Starting Price

$0/Month

Key Features

Chain-of-Causation reasoning traces for interpretable decisions
Diffusion-based trajectory decoder for precise 6.4s future paths
Multimodal input: multi-camera video, text commands, egomotion history
State-of-the-art performance on long-tail autonomous driving scenarios
Open inference code and non-commercial model weights

Pros

First open reasoning VLA model for autonomous vehicles
Provides human-like explainable decisions alongside actions
Strong benchmarks in open-loop and closed-loop evaluations
Supports research and fine-tuning for AV development
Runs on consumer-grade high-end GPUs with 24GB+ VRAM

Cons

Model weights restricted to non-commercial use
Requires powerful NVIDIA GPU for inference
Focused only on autonomous driving research
No hosted demo or easy web interface

Alpamayo is an excellent choice for researchers, autonomous vehicle developers, and AI engineers working on explainable, reasoning-based self-driving systems, offering a breakthrough in bridging perception, reasoning, and action.

FAQs

What is Alpamayo?
Alpamayo is NVIDIA’s 10B parameter vision-language-action model that enables autonomous vehicles to reason step-by-step (via Chain-of-Causation) before predicting safe driving trajectories, making decisions more interpretable and robust in complex scenarios.
Is Alpamayo completely open-source?
Model weights are available for non-commercial research use on Hugging Face, while the inference code is fully open under Apache 2.0 license.
When was Alpamayo released?
The model was initially uploaded to Hugging Face on December 3, 2025, with major public announcement and renaming to Alpamayo 1 during CES 2026 in early January.
What hardware do I need to run Alpamayo locally?
You need at least one NVIDIA GPU with 24GB+ VRAM (such as RTX 4090 or H100) and Linux OS, along with PyTorch, Transformers, and DeepSpeed for optimal inference performance.