SemiconductorsBRIEF

NVIDIA Opens Alpamayo 2 Super for Commercial Use, Says Model Leads Every Autonomous-Driving Benchmark It Tested

N
NathanTechnology Editor · Technical Lead
Published · Updated
According to NVIDIA's official blog, the company released Alpamayo 2 Super — a model built on NVIDIA Cosmos 3 Super Reasoner — on Hugging Face under the OpenMDW-1.1 license, permitting commercial fine-tuning and redistribution. NVIDIA said the model ranks first on the LingoQA benchmark among nearly 40 models, and the wider Alpamayo family has surpassed 500,000 Hugging Face downloads.

Technical Specifications and Model Iteration

According to NVIDIA's blog post announcing the release, Alpamayo 2 Super is "built on NVIDIA Cosmos 3 Super Reasoner and post-trained with reinforcement learning." NVIDIA said the model "advances the AV ecosystem on two fronts: open commercial licensing and leading multitask capabilities for autonomous driving" — a characterization from the company's own announcement rather than an independent assessment.

On scale, NVIDIA stated that Alpamayo 2 Super "offers 3x the scale of the 10-billion-parameter NVIDIA Alpamayo 1.5 and Alpamayo 1 models," marking it as the largest model in the Alpamayo lineup to date, though NVIDIA did not disclose an absolute parameter count for the new model.

Open Licensing and Commercial Deployment: Hugging Face Release and the OpenMDW License

NVIDIA announced that Alpamayo 2 Super is "available now for commercial use" as part of what the company calls "the most-adopted open reasoning models for autonomous driving on Hugging Face." The model is distributed under OpenMDW-1.1, which NVIDIA describes as "the Linux Foundation's permissive license for open AI model distributions," covering "fine-tuning, derivative models and commercial redistribution."

NVIDIA further stated that this licensing approach now extends beyond the single model: "The OpenMDW license is now being applied across the entire Alpamayo model family so developers can deploy any of the models commercially without requiring additional permissions." This means earlier releases in the family are covered by the same commercial terms as the new Alpamayo 2 Super.

Benchmark Performance: Vision-Language Understanding and Autonomous-Driving Evaluations

NVIDIA reported that Alpamayo 2 Super "ranks first on LingoQA, an autonomous driving reasoning benchmark, among nearly 40 models evaluated." In NVIDIA's own testing using the Lingo-Judge metric, the company said the model outperformed three named competitors by specific margins:

Comparison ModelAlpamayo 2 Super's Margin (Lingo-Judge, per NVIDIA testing)
Qwen2.5-VL 72B+17.0 points
Gemini 2.5 Pro+15.1 points
GPT-4o+23.2 points

NVIDIA said these results demonstrate what it called "state-of-the-art reasoning for driving-centric scenarios." The company also stated that "Alpamayo 2 Super also ranks first across all autonomous driving benchmarks evaluated by NVIDIA," a claim NVIDIA said "underscor[es] its leading performance across a broad range of AV capabilities." All of these benchmark results come from NVIDIA's own testing, as disclosed in its announcement.

Purpose-Built Architecture: Five Coupled Outputs and Automated Training Data

Per NVIDIA's description, Alpamayo 2 Super produces five tightly coupled outputs for each driving situation: a trajectory describing the vehicle's planned path; a chain-of-causation (CoC) trace explaining the reasoning behind a decision; a meta-action (such as yield, lane change, or stop) capturing the model's intent; reasoning auto-labels that generate CoC annotations for training and validation data; and visual question-answering responses with 2D visual grounding linking answers to specific regions in camera images.

NVIDIA said this grounding capability allows the model to function as an automated annotator: "By linking its reasoning to specific regions in camera images, the model can transform raw driving clips into richer training data, compressing annotation cycles from months to days."

Safety Design: ISO/PAS 8800 Alignment and Halos Validation

Building on the chain-of-causation trace described above, NVIDIA stated that these "CoC traces integrate with NVIDIA Halos safety-validation workflows and support AI safety aligned with ISO/PAS 8800 requirements, providing a stronger foundation for AV safety engineering." This connects the model's reasoning-output architecture (the CoC trace) directly to a named industry safety standard and NVIDIA's own Halos validation tooling.

Open Ecosystem, Supporting Tools, and Adoption Metrics

NVIDIA listed three components supporting the Alpamayo open ecosystem: "NVIDIA AlpaSim, which provides closed-loop simulation. NVIDIA AlpaGym, which enables high-throughput reinforcement learning. NVIDIA Physical AI Open Datasets, which supply data for training and testing."

On adoption, NVIDIA reported that "Alpamayo has already surpassed 500,000 downloads on Hugging Face, reinforcing its position as the most-adopted open reasoning model family for autonomous driving on the platform." That download figure applies to the Alpamayo family as a whole, not to Alpamayo 2 Super alone.

What This Means

Taken together, the facts NVIDIA disclosed describe a layered release: a commercially unrestricted license (OpenMDW-1.1) now applied retroactively across the entire Alpamayo family, paired with benchmark claims — first on LingoQA among nearly 40 models, and first across every NVIDIA-evaluated AV benchmark — that come exclusively from NVIDIA's own internal testing rather than a third-party leaderboard. The model's chain-of-causation output is positioned both as a safety-engineering artifact tied to ISO/PAS 8800 and Halos validation, and as a data-generation tool that NVIDIA says shortens annotation cycles from months to days. The 500,000-download figure for the broader Alpamayo family is the only independent-platform signal in NVIDIA's announcement; the performance and safety claims rest on NVIDIA's own reporting.

📊 Evidence

FAQ

What license governs commercial use of Alpamayo 2 Super?

According to NVIDIA, the model is released under OpenMDW-1.1, the Linux Foundation's permissive license, which covers fine-tuning, derivative models, and commercial redistribution. NVIDIA said this license now applies across the entire Alpamayo model family.

How does Alpamayo 2 Super compare to other vision-language models on driving benchmarks?

In NVIDIA's own Lingo-Judge testing, the company reported Alpamayo 2 Super outperformed Qwen2.5-VL 72B by 17.0 points, Gemini 2.5 Pro by 15.1 points, and GPT-4o by 23.2 points, and said it ranks first on the LingoQA benchmark among nearly 40 models evaluated.

What outputs does Alpamayo 2 Super generate for each driving scenario?

NVIDIA said the model produces five coupled outputs: a trajectory, a chain-of-causation reasoning trace, a meta-action, reasoning auto-labels for training data, and visual question-answering responses with 2D visual grounding.

How many downloads has the Alpamayo family reached?

NVIDIA reported the Alpamayo model family has surpassed 500,000 downloads on Hugging Face, which it described as making it the most-adopted open reasoning model family for autonomous driving on the platform.

📎 Sources

  1. blogs.nvidia.com
N
NathanTechnology Editor · Technical Lead

Related

BRIEF

At SIGGRAPH 2026, NVIDIA Announces Cosmos 3 Edge, DGX Station and MotionBricks for Agentic and Physical AI

According to NVIDIA's SIGGRAPH 2026 blog post, the company announced the 4-billion-parameter Cosmos 3 Edge model, DGX Station systems built on the GB300 Grace Blackwell Ultra Desktop Superchip with up to 20 petaflops of FP4 compute, a Synthetic Video Detector microservice reaching 92% accuracy on uncompressed video, and MCP-based AI agent integrations across five creative-software platforms.

Nathan ·
BRIEF

Intel's New Mexico Fab Pushes Packaging Reticle Limit to 8x Today, Targets 12x+ by 2028

According to Intel's newsroom announcement, Intel's Fab 9 advanced packaging site in Rio Rancho, New Mexico has scaled its 'reticle limit' to 8 times the current industry standard, with a target of over 12 times by 2028. Intel says the facility grew from 25 employees in 1980 to 2,700 employees and 500 suppliers today, using Foveros and EMIB/EMIB-T technologies to package AI chips.

Nathan ·