NVIDIA releases Alpamayo 2 Super, an open frontier model for robotaxis, for commercial use

NVIDIA made Alpamayo 2 Super, its frontier open-weights reasoning model for robotaxis and autonomous vehicles, available for commercial use on August 4, 2026, under a license that permits fine-tuning, derivative models, and commercial redistribution, according to NVIDIA's official announcement.

What's new

Alpamayo 2 Super is part of what NVIDIA calls "the Alpamayo family, the most-adopted open reasoning models for autonomous driving." For a given driving situation, the model produces five coupled outputs:

• A trajectory describing the vehicle's planned path
• A chain-of-causation (CoC) trace that explains the reasoning behind the decision
• A meta-action (e.g., yield, lane changes, stops) capturing the model's intent
• Reasoning auto-labels usable as training data
• Visual question-answering responses with 2D visual grounding

NVIDIA also lists scene understanding, model critiquing, and knowledge distillation among its capabilities. The model is released as open weights: "Open weights make that path economically viable" for companies building and adapting commercial autonomous-driving stacks. It ships under the "OpenMDW-1.1" license, which NVIDIA says explicitly covers "fine-tuning, derivative models and commercial redistribution." The model is hosted on Hugging Face, where NVIDIA says the Alpamayo family has "surpassed 500,000 downloads."

Context

Autonomous-vehicle perception and planning have historically relied on closed, in-house models built by each robotaxi or AV company individually — an expensive, duplicated effort across the industry. Alpamayo's open-weights approach, paired with a chain-of-causation trace that explains why the model chose a given trajectory, targets both the cost problem and the interpretability problem that has slowed AV deployment and regulatory trust.

This release follows earlier Alpamayo models from NVIDIA and represents the "Super" tier of the family — positioned as the most capable variant, per the "frontier" framing NVIDIA uses in the announcement.

Why it matters

Making a frontier-capability autonomous-driving reasoning model available under a commercial-use, open-weights license lowers the barrier for smaller AV and robotaxi developers to build on state-of-the-art perception-and-planning technology rather than developing it from scratch — a meaningful shift in a field where the leading closed models have been a significant competitive moat. The chain-of-causation output in particular addresses a persistent criticism of AV decision-making models: that they act as black boxes. A model that explains its reasoning trace alongside its trajectory gives downstream developers and, potentially, regulators a concrete artifact to audit decisions after the fact, which could matter as autonomous-vehicle oversight tightens.