Nemotron 3.5 Super VL Reaches A Score Of 26 With Low-Cost Tasks

🔍 Read the full analysis: Nemotron 3.5 Super VL Reaches A Score Of 26 With Low-Cost Tasks on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on monitors, keyboards and dev gear

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

NVIDIA released Nemotron 3.5 Super VL on October 15, with an Artificial Analysis Intelligence Index score of 26 and a reported cost of $0.25 per index task. The 120-billion-parameter model, with 12 billion active parameters, combines text and image input with a commercial-use license; its appeal is speed and deployment flexibility, though the score trails several alternatives that cost the same or less.

Artificial Analysis places the model at 26.5 on its index, compared with 13 for its predecessor, Nemotron 3 Super. It ranks first among 67 open-weight models in the stated 40-billion-to-150-billion-parameter class, but around 18th when compared across all open-weight models. The size-class ranking is not a claim that it is the strongest open model overall.

The reported API prices are $0.20 per million input tokens and $0.575 per million output tokens, with cached input priced at $0.055 per million tokens. Artificial Analysis estimates one index task costs $0.25, versus $1.64 for Nemotron 3 Super. Its launch write-up attributes much of the lower task cost to reduced output: about 46,000 output tokens per index task, compared with 114,000 for the predecessor.

Speed varies by measurement. A pre-release DeepInfra endpoint serving NVFP4 weights produced about 630 output tokens per second on prompts of 100,000 tokens. Artificial Analysis’ model-page median across six providers was 307 tokens per second on those long prompts, with a reported range from 106 at the fifth percentile to 534 at the 95th percentile. Median first-token latency was 0.89 seconds, but the source reports about 11 seconds before the first answer token when the model reasons before responding.

Speed Comes With Trade-Offs

For teams running high-volume document or repository workflows, the relatively low estimated task cost and support for long prompts could make Super VL worth testing. The price breakdown matters: Artificial Analysis estimates that roughly 89% of the $0.25 task cost is input, including 69% attributed to cache writes. Actual spending will depend on how much material a system sends and whether it can reuse cached input, not simply on the output-token rate.

Raw intelligence-per-dollar is less compelling. The source compares Super VL’s score of 26 and $0.25 task cost with MiMo-V2.6-Pro at 46 and $0.13, MiMo-V2.6-Flash at 38 and $0.06, and Inkling Small at 26 and $0.09. These comparisons use Artificial Analysis’ reported figures, but do not by themselves establish which model will perform best on a particular team’s workload.

Super VL’s case rests on a different combination: throughput, low reported latency, commercial-use terms, and the option to run weights on NVIDIA hardware. Those factors can matter for organizations with infrastructure, data-control, or licensing requirements. They should be weighed against task quality and measured costs in the intended deployment, rather than treating the index score as a complete purchasing guide.

A Larger Score Than Its Predecessor

The launch follows Nemotron 3 Super, which the source lists at 13 on the Artificial Analysis Intelligence Index. The new model’s reported score is nearly twice that figure, while its estimated cost per index task drops from $1.64 to $0.25. These are Artificial Analysis results and cost estimates, not a guarantee of performance or billing for every provider and workload.

The model is described as a 120-billion-parameter mixture-of-experts system with 12 billion active parameters and a one-million-token context window. It accepts text and images and produces text. Its weights were released in NVFP4, a four-bit format, and the source lists the OpenMDW-1.1 license as allowing commercial use.

The source also reports gains on selected evaluations: GDPval-AA, covering economically useful tasks, rose to 1,162 Elo from 490 for Nemotron 3 Super; AA-Briefcase reached 999 Elo, compared with 876 for Nemotron 3 Ultra. AutomationBench-AA was 14.8%, versus 3.0% for Ultra. On AA-Omniscience, the reported hallucination rate fell from 87% to 36%. These are benchmark-specific results, not proof of equivalent gains across real-world applications.

“Super VL is not the best value for raw intelligence, even among open models.”

— Thorsten Meyer, ThorstenMeyerAI.com

Provider Results Still Vary

The available figures do not establish how the model will perform across all providers or production workloads. The 630-token-per-second result came from one pre-release endpoint; the six-provider median for 100,000-token prompts was 307, with a wide reported spread. Provider hardware, serving settings, prompt length, and workload may affect measured speed and latency.

The $0.25 figure is an estimate for an Artificial Analysis index task, not a fixed charge for every use. The supplied material does not give enough detail to generalize the estimate to different prompt sizes, caching patterns, or production traffic. Nor does a benchmark score settle whether the model’s quality is sufficient for a specific task.

Licensing also requires careful reading. The source describes OpenMDW-1.1 as permitting commercial use and says Super VL leads among permissively licensed U.S. open-weight models. It contrasts this with Muse Spark Open’s more restrictive Muse Model License. The material does not provide the full license text or legal analysis, so organizations should review the terms for their own use.

Test Against Real Workloads

The next useful step for prospective users is to test the model on their own prompts and compare quality, total cost, and response time with alternatives. Those tests should use the provider or hardware setup intended for deployment, especially for long-context tasks where reported throughput varies between endpoints.

Teams should also check the full license and confirm how their input data, caching, and self-hosting arrangements will be handled. The supplied source gives no announced schedule for further model updates or additional benchmark releases. For now, the key evidence is the launch-day evaluation, the reported pricing, and provider-specific performance that remains subject to verification in production.

Key Questions

What score did Nemotron 3.5 Super VL receive?

Artificial Analysis reported an Intelligence Index score of 26.5, commonly rounded to 26. Its source places the model first among 67 open-weight models in the 40-billion-to-150-billion-parameter size class, not first among all open models.

How much does an index task cost?

Artificial Analysis estimates $0.25 per index task. That is a benchmark-specific estimate; actual costs depend on input volume, caching, provider pricing, and workload.

How fast is the model?

A pre-release DeepInfra endpoint reached about 630 output tokens per second on 100,000-token prompts. The median across six providers was 307 tokens per second for the same prompt length, so users should benchmark their intended endpoint.

Can businesses use Nemotron 3.5 Super VL?

The source lists its OpenMDW-1.1 license as allowing commercial use. Businesses should review the full license terms and their deployment requirements before relying on that description.

Is it the best model for the price?

Not according to the comparisons in the source: several listed models score higher at similar or lower estimated task costs. Super VL’s reported advantages instead include speed, latency, licensing, and self-hosting options; which model is best depends on the workload and deployment needs.

Source: ThorstenMeyerAI.com