Ethical Stack

Nemotron 3.5 family

Open model family for agent workflows. Latest verified general-purpose release in this pass is Nemotron 3.5 Super VL (120B) alongside Bolt edge variants, announced October 15, 2026.

Score calculation

(7 + 7) ÷ 2 = 7.0

2 of six dimensions rated. Unrated dimensions and BDS status are excluded.

Full scoring method ↗
Setup details

NVIDIA's releases provide varied openness and efficiency claims; these are model family claims and do not establish use by a particular customer or military/Israel-related deployment.

NVIDIA announced Nemotron 3.5 Super VL and Bolt variants on October 15, 2026. The E4B model was scheduled for October 23; publication/availability of that exact variant was not independently confirmed in the release source.

Energy

↑

NVIDIA frames Lightning and Bolt around efficiency and local deployment; it publishes throughput/size claims, not model-specific energy, water, or emissions measurements. Faster throughput is not an energy-per-task result.

Reasoning & sources

NVIDIA’s announcements give throughput, model size, and local deployment claims but no electricity or emissions measurements. Such figures can help estimate capacity but cannot establish energy per useful task without hardware, utilization, and quality measurements.

No independently measured power, water, or emissions figures for the family.

Training rights

↑

NVIDIA says some Nemotron releases include open datasets and training recipes; the October Super VL/Bolt announcement emphasizes open models but this page does not enumerate all data rights. Check the exact checkpoint card and license.

Reasoning & sources

NVIDIA says some Nemotron releases include datasets and recipes, while the cited 3.5 announcements do not enumerate the exact family’s corpus rights. Openness varies by release artifact; inspect each checkpoint’s dataset and license rather than generalizing from the family label.

Open data/recipe completeness differs by release; exact Super VL corpus rights need verification.

Privacy

↑

Open weights permit local deployment, where prompt handling is controlled by the operator; managed NIM/cloud use has separate provider terms. The model release does not define all downstream privacy practices.

Reasoning & sources

Open weights permit local hosting where the operator controls prompts, while NIM and cloud deployments introduce provider-specific handling. This privacy benefit applies only to deployments where the operator manages storage, telemetry, and access.

Open weights support local deployment and operator control, though hosted providers have separate terms.

Need deployment-specific logging, retention, and data-training terms.

Openness

↑

NVIDIA describes Super VL and Bolt as open models and says users can post-train and deploy them; licenses, datasets, and degree of full-stack reproducibility should be checked per artifact.

Reasoning & sources

NVIDIA describes the relevant models as open and supports post-training and deployment, but openness depends on the individual checkpoint’s license and accompanying datasets. The available weights and customization support reuse, while incomplete training data and recipes limit reproduction.

NVIDIA describes open models and customization; exact artifact license/data completeness varies by release.

Not all family releases necessarily share the same license or data openness.

Worker treatment

↑

NVIDIA’s Nemotron 3.5 Content Safety card says its prompt, image, and response data were human-annotated through SuperAnnotate, with synthetic data and LLM judges also used. It does not publish annotator pay or working conditions.

Reasoning & sources

The model card identifies a named human-annotation vendor and says annotators were instructed to skip very graphic images. This gives concrete evidence of human work and one exposure control, while leaving compensation and broader protections unanswered.

The Content Safety card does not establish pay, working conditions or protections across every Nemotron 3.5 variant. Current vendor-level outcomes remain unverified.

Government & military

↑

NVIDIA says Palantir uses Nemotron open models for U.S. government agencies and says Nemotron models are available through AWS GovCloud at DoD Impact Levels 2, 4, and 5. These company materials do not name Nemotron 3.5 as a military deployment.

Reasoning & sources

NVIDIA’s federal page documents Nemotron in government deployments and DoD-cleared hosting, and its Palantir announcement names Nemotron open models serving U.S. agencies. The pages do not identify the 3.5 checkpoint or a particular military mission, so they establish family-level government use rather than a 3.5-specific contract.

Which Nemotron releases are deployed, whether a specific use is military rather than civilian government work, and the actual mission and safeguards remain unestablished.

BDS status

↑

We did not find NVIDIA among the named targets in the two official BNC campaign lists. This is a limited list check, not BDS approval.

Reasoning & sources

NVIDIA is not a named target in the checked lists, but BDS uses selective campaigns, so absence cannot establish a clean record. Running downloaded weights locally is a different decision from paying the developer or a cloud host; check the host as well.

This review covers the named targets in the BNC consumer/pressure guide and No Tech campaign, not every BDS campaign, local group, investment or supplier. Lists and ownership can change. A company’s absence from these lists is not an endorsement or proof that it has no relevant relationships.

All sources

Side by side