Can Nemotron 3.5’S Score Narrow America’s Open-Weight Gap?

🔍 Read the full analysis: Can Nemotron 3.5’S Score Narrow America’s Open-Weight Gap? on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

NVIDIA released Nemotron 3.5 Super VL and Nemotron 3.5 Bolt 74B-A9B on Oct. 15, and independent evaluator Artificial Analysis scored them at 26 and 21, respectively, on its Intelligence Index. The results strengthen NVIDIA’s position among open-weight models that allow commercial use, but Super VL remains behind the highest-scoring models in the broader rankings, where Chinese developers dominate.

Super VL is the larger model, with 120 billion total parameters and 12 billion active per token. It accepts text and images and returns text, has a reported one-million-token context window, and uses the OpenMDW-1.1 license, which the source says allows commercial use. Artificial Analysis recorded a score of 26, or 26.5 before rounding, and a cost of $0.25 per index task.

Compared with its predecessor, Nemotron 3 Super, Super VL’s Intelligence Index score rose from 13 to 26 in seven months, while reported output tokens per index task fell from about 114,000 to about 49,000. The source gives the predecessor’s task cost as $1.64, compared with $0.25 for Super VL. These are evaluator-reported figures, not a guarantee of costs or performance in every deployment.

Bolt is a smaller mixture-of-experts model with 74 billion total parameters, nine billion active, and a 256,000-token context window. It scored 21. Artificial Analysis placed it fourth among models under 100 billion total parameters and first among U.S. models in that size class. The source says Bolt has no public API yet, leaving its hosted price unreported.

A stronger U.S. model, not a new leader

The release matters for developers choosing models they can inspect, adapt, or host, particularly when commercial licensing and operating cost are important. Super VL’s reported score and per-task cost suggest a sizable improvement over NVIDIA’s earlier model, while Bolt offers a lower-parameter option aimed at less demanding hardware. Actual results will still depend on deployment, serving provider, and workload.

But the results do not establish that the United States has caught up in open-weight model performance. The source’s Artificial Analysis snapshot places Super VL at 18th among open-weight models overall and 11th among permissively licensed models. The strongest listed model scores 46, 20 points above Super VL. The benchmark is one way to compare models, not a full measure of their usefulness, reliability, or suitability for a particular task.

How the rankings compare

Artificial Analysis published its evaluation on the day NVIDIA released the models. Its Intelligence Index combines results across evaluations; the supplied source reports the scores and selected task-level comparisons but does not provide the full methodology or every underlying result. Rankings should be read as a dated benchmark snapshot, not a permanent ordering.

The source says nine of the ten highest-scoring open-weight models in its listing are from Chinese labs. It identifies China’s MiMo-V2.6-Pro at 46, several other Chinese models at 44 to 45, and the UAE’s K2 Horizon at 31. Meta’s Muse Spark Open, released on the same day as NVIDIA’s models, scored 39. Its Muse Model License has commercial restrictions, according to the source, unlike Super VL’s commercially permissive OpenMDW-1.1 license.

Hardware efficiency is another part of NVIDIA’s case. Artificial Analysis found that Super VL’s NVFP4 version scored within 0.1 points of its full-precision BF16 weights, with no individual evaluation moving by more than 2.6 points. The source also reports a 630-output-token-per-second result from a pre-release DeepInfra endpoint using 100,000-token prompts; the median across six listed providers was 196 tokens per second. Those figures indicate provider-specific variation, not a speed every user should expect.

Pricing and deployment remain unsettled

The source does not give a public API price for Bolt, so its cost per index task cannot yet be compared with Super VL or other hosted models. It also says Artificial Analysis had no license listing for Bolt at the time of its evaluation. Prospective users should check the model’s published terms before commercial deployment.

Some performance figures are tied to particular conditions: the 630-token-per-second result came from a pre-release endpoint, while provider measurements varied substantially. The supplied material does not include the full benchmark methodology, model cards, or details of all evaluation runs. The rankings and cost figures therefore do not by themselves settle how either model will perform across real-world workloads.

The source also says Muse Spark Open scored 39 but carries commercial restrictions. The supplied text cuts off before setting out the license’s specific terms, so the exact limits cannot be described here. More broadly, the benchmark snapshot does not show whether the relative rankings will hold as providers update models or evaluators add results.

Watch for public access and follow-up tests

The next practical milestone for Bolt is the arrival of a public API and a published license, which would allow buyers to evaluate hosted pricing and commercial-use terms. For Super VL, provider-level results and testing on users’ own workloads will show how closely the reported costs and speed translate into practice.

Further Artificial Analysis evaluations could also change the ranking as more models and versions are added. Until then, the confirmed takeaway is narrower: NVIDIA’s new releases improve its benchmark position, especially against its own predecessor, while the supplied rankings continue to place several Chinese models and Meta’s restricted-license Muse Spark Open ahead of Super VL.

Key Questions

What did Nemotron 3.5 Super VL score?

Artificial Analysis gave Super VL an Intelligence Index score of 26, with an unrounded score of 26.5 reported in the source.

Did Super VL close the gap with the leading open-weight models?

No. The source places it 18th in the overall open-weight ranking, while the highest listed model scored 46. Super VL’s result improves NVIDIA’s position but does not make it the overall leader.

How does Super VL compare with Nemotron 3 Super?

Its score rose from 13 to 26. The source also reports a reduction in output tokens per index task from about 114,000 to 49,000 and in cost per task from $1.64 to $0.25.

Can businesses use Super VL commercially?

The source says Super VL uses the OpenMDW-1.1 license, which allows commercial use. Users should consult the actual license terms before deployment.

Is Bolt available through an API, and what does it cost?

The source says Bolt had no public API at the time of the evaluation and that no API price was available. Its eventual hosted cost remains unknown.

Source: ThorstenMeyerAI.com