How Much Does A Nemotron 3.5 Super VL Task Cost?

🔍 Read the full analysis: How Much Does A Nemotron 3.5 Super VL Task Cost? 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’s Nemotron 3.5 Super VL costs about $0.25 per Artificial Analysis Intelligence Index task, according to figures published after the model’s October 15 release. Its API rates are $0.20 per million input tokens and $0.575 per million output tokens; the model’s speed and commercial-use license may appeal to operators, but the cited comparisons show that several alternatives deliver higher benchmark scores for similar or lower task costs.

Artificial Analysis assigns Super VL an Intelligence Index score of 26.5 and reports a task cost of $0.25. The source analysis rounds the score to 26 in its headline. It says the model leads a size class of open-weight models with 40 billion to 150 billion total parameters, but ranks about 18th when compared with open-weight models across all size classes. The model has 120 billion total parameters and 12 billion active in a mixture-of-experts design, and accepts text and images.

The listed API prices are $0.20 per million input tokens and $0.575 per million output tokens. Cached input is listed at $0.055 per million tokens, a 73% reduction from the uncached input price. Artificial Analysis’s task-cost breakdown, as described in the source, puts about 89% of the $0.25 total on input, with cache writes alone accounting for 69%; output accounts for less than three cents. The cost of a particular deployment will depend on its prompt, caching and output volume, so the benchmark task figure is not a universal per-request price.

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, according to the source. Across six providers, the reported median was 307 tokens per second at that prompt length, with a range from 106 to 534. Median first-token latency was 0.89 seconds, while the source says the model’s reasoning phase can take about 11 seconds before the first answer token and a 500-token answer takes about 13.6 seconds end to end.

Cost Depends on the Workload

The headline $0.25 task price is useful for comparing models under Artificial Analysis’s evaluation, but it should not be treated as a fixed bill for every application. The source’s cost breakdown puts most spending on input, particularly cache writes. Teams that repeatedly send long documents or code repositories may face a very different bill from those using short prompts, even though both use the same published token rates.

The value decision is not just about price per task. In the comparisons supplied by the source, MiMo-V2.6-Pro scores 46 at $0.13 per task, while Inkling Small scores 26 at $0.09. Other cited alternatives include MiMo-V2.6-Flash at 38 for $0.06 and DeepSeek V4.1 Flash at 39 for $0.27. Those figures suggest Super VL is not the cheapest route to a given score in this benchmark. Its case instead rests on factors such as measured speed, commercial licensing, NVIDIA hardware compatibility and the option to host the weights locally.

For operators, that distinction matters. A team prioritizing a high benchmark score per dollar may prefer another model, while a team with strict infrastructure or licensing requirements may value deployment control more. The benchmark and cost figures provide one comparison point, not a complete measure of production performance or total ownership cost.

From Release to Cost Comparison

NVIDIA released Nemotron 3.5 Super VL on October 15, 2026. Artificial Analysis published benchmark figures the same day, according to the source analysis dated October 16. The model is released in NVFP4, a four-bit format, and its listed context window is one million tokens. Its stated license, OpenMDW-1.1, allows commercial use, although prospective users should review the license terms for their own deployment.

The source compares its $0.25 task cost with $1.64 for Nemotron 3 Super, its predecessor. It also reports that Super VL produced about 46,000 output tokens per Intelligence Index task, compared with 114,000 for the predecessor. The analysis characterizes the newer model as twice the predecessor’s score at roughly 85% lower task cost. Those are benchmark-specific comparisons, not guarantees that every production workload will see the same savings.

The claim that Super VL is the top-scoring permissively licensed U.S. open-weight model also has a defined scope. The source says its score is highest within that category, while noting that Meta’s Muse Spark Open scored 39 but uses a license with commercial restrictions. The result should not be read as a lead over every open-weight model: the source places Super VL around 18th across all open-weight models.

““Super VL is a very good engine and an average brain.””

— Thorsten Meyer, ThorstenMeyerAI.com

Real-World Costs and Speeds

The $0.25 figure describes an Artificial Analysis evaluation task; it does not establish what a typical customer request or production workflow will cost. Actual spending depends on token use, caching, provider rates and workload design. The source does not provide a representative range of production bills or a cost comparison based on independently observed customer deployments.

The fastest reported result came from one pre-release DeepInfra endpoint. The source also reports lower median speed across six providers and a wide provider range, so the 630-token-per-second result should not be assumed for every endpoint. The supplied material does not specify how final production configurations, hardware availability or later provider changes could affect those results.

The benchmark provides evidence about selected evaluation tasks, not a guarantee of performance across all applications. The source reports gains in knowledge-work and SaaS-automation tests and a lower hallucination rate than the predecessor, but those figures do not establish error rates for a particular organization’s documents, users or risk settings. The source excerpt ends as it begins to describe additional evaluation results, so it does not provide a complete account of the model’s limitations.

Test the Intended Deployment

Teams considering Super VL should measure it on the provider and hardware configuration they intend to use, rather than relying on the single-endpoint peak. A useful comparison would track input and output token volume, cache use, time to first answer, end-to-end latency and task accuracy on the organization’s own workloads. Those measurements can show whether the model’s speed and hosting options justify its cost relative to alternatives.

Users should also check current provider pricing and the OpenMDW-1.1 license terms before committing to a commercial or self-hosted deployment. The source lists six providers but does not give a full provider-by-provider price and performance table. Until those details and independent production results are available, the reported benchmark task price is best treated as a reference point rather than a final operating-cost estimate.

Key Questions

How much does one Nemotron 3.5 Super VL task cost?

Artificial Analysis lists the cost of one Intelligence Index task at about $0.25. That is a benchmark-specific estimate, not a fixed price for every user request.

What are the model’s API token rates?

The reported rates are $0.20 per million input tokens and $0.575 per million output tokens. Cached input is listed at $0.055 per million tokens.

Why can costs vary between users?

Costs depend on the number of input and output tokens and on cache use. In the source’s benchmark breakdown, input-related charges account for about 89% of the task cost, so long or repeatedly supplied prompts can weigh heavily on the bill.

Is Nemotron 3.5 Super VL the best-value model?

Not on the cited Intelligence Index score per task alone. The source lists alternatives with higher scores at similar or lower costs, including MiMo-V2.6-Pro and MiMo-V2.6-Flash. Super VL may still suit teams that prioritize speed, commercial licensing or self-hosting.

How fast is Nemotron 3.5 Super VL?

A pre-release DeepInfra endpoint reached about 630 output tokens per second on 100,000-token prompts, but the reported median across six providers was 307 tokens per second for that prompt length. Performance depends on the provider and configuration.

Source: ThorstenMeyerAI.com