NVIDIA Alpamayo

NVIDIA Alpamayo

Open VLA models, frameworks, and datasets for reasoning-based autonomous vehicles.

Overview

What Is NVIDIA Alpamayo?

NVIDIA Alpamayo, winner of the COMPUTEX Best Choice Award,  is a family of open VLA models, simulation frameworks, reinforcement learning infrastructure, and physical AI datasets, designed to accelerate the development of safe, transparent, and reasoning-based robotaxis & autonomous vehicles (AVs).

Built to accelerate Level 4 autonomy, Alpamayo lets vehicles perceive, reason, and act with human-like judgment. By reasoning through rare and complex scenarios rather than relying on predefined rules, Alpamayo advances safer, more transparent autonomous driving at scale.

 

NVIDIA Alpamayo 2 Super

Alpamayo 2 Super is a 32 billion-parameter reasoning‑based vision language action (VLA) model. It extends the NVIDIA Alpamayo family of open AI models, simulation frameworks, and physical AI datasets for safe robotaxi (Level 4) development.

Post-Train Autonomous Vehicle Models in Closed-Loop with NVIDIA Alpamayo

This blog walks users through post-training AV models with closed-loop reinforcement learning using AlpaGym, part of the NVIDIA Alpamayo family.

Technology

Accelerated Autonomous Vehicle Development

See how Alpamayo is used across the AV development process.

Data and Labeling

Physical AI open datasets provide real-world driving data. Alpamayo open VLA models generate Chain-of-Causation (CoC) reasoning traces at scale.

Model Training and
Post-training

Alpamayo open VLA models are trained and optimized with AlpaGym through closed-loop RL.

Simulation and Validation

AlpaSim validates model behavior in closed-loop simulation before real-world deployment.

In-Vehicle Deployment

Alpamayo enables a validated path to in-vehicle deployment on NVIDIA DRIVE AGX Thor™.

Demo

Building Autonomous Vehicles That Reason

See how Alpamayo enables a vehicle to reason through complex, real-world scenarios and explain every decision.

Benefits

What Does Alpamayo Bring to Autonomous Vehicle Development?

Open and Transparent AI

Alpamayo delivers fully open models, simulation frameworks, and datasets that promote transparency, reproducibility, and trust. Developers can inspect, extend, and fine-tune the technology to meet regional safety standards and regulatory requirements.

Reasoning-Based Autonomy

Unlike perception-only systems, Alpamayo empowers vehicles to explain why they act. Vision-language-action models reason over complex driving scenes, verbalize decision logic, and support interpretable, auditable autonomy.

High-Fidelity Simulation at Scale

With open simulation frameworks and neural reconstruction tools, Alpamayo enables closed-loop testing across diverse traffic scenarios, weather conditions, and edge cases. This greatly reduces real-world testing time and accelerates validation.

An Accelerated Path to Level 4

By standardizing foundational autonomy components, Alpamayo helps OEMs, Tier 1s, startups, and researchers shorten development cycles, align with global safety expectations, and focus on differentiation rather than rebuilding core AI systems.

Models and Frameworks

Core Components of the Alpamayo Portfolio

Alpamayo : Open Reasoning VLA Model

An open reasoning VLA model for autonomous driving, Alpamayo is built on NVIDIA Cosmos™ and processes multi-camera video, navigation inputs, and driving context to generate trajectories and Chain-of-Causation reasoning traces. It’s available in 10 B parameters with Alpamayo 1 Nano and 1.5 Nano, and scales to 32 B with Alpamayo 2 Super.

AlpaSim: Open Simulation Framework

An open-source closed-loop AV simulation framework, AlpaSim validates model decisions against real-world consequences, accelerating safety validation without scaling physical test fleets.

AlpaGym: Closed-Loop Reinforcement Learning

NVIDIA AlpaGym is a high-throughput closed-loop RL training framework, training AV models through continuous decision and observation cycles, producing models validated against the full complexity of real-world driving.

Physical AI Open Datasets: Real-World Data

The largest open multi-sensor driving dataset available, Physical AI Open Datasets provide real-world driving data with Chain-of-Causation reasoning labels across 25 countries, purpose-built for long-tail scenario coverage.

Resources

Explore the Latest in Automotive

In-Car Reasoning: AI That Drives, Explains Decisions, and Responds to Passengers

This video showcases four capabilities of NVIDIA Alpamayo: chain-of-thought, verbalized reasoning, natural language Q&A, and language instructions and commands.

How Autonomous Vehicles Learn to Reason With NVIDIA Alpamayo

See how open NVIDIA Alpamayo models, simulation frameworks, and physical AI datasets accelerate reasoning-based development for autonomous vehicles and self-driving cars.

Next Steps

Build with Alpamayo

Use the open models, frameworks, and datasets to take NVIDIA Alpamayo from research to real-world autonomous driving.

Get in Touch

Connect with NVIDIA to learn more about building autonomous vehicles.

Stay Up to Date on NVIDIA Automotive News

Get the latest automotive news, technologies, breakthroughs, and more sent straight to your inbox.

Contact Us

This form is for automotive inquires only. For other inquiries, please click here.