Why Look At Nemotron 3.5 Super VL’s 25-Cent Task Price?

🔍 Read the full analysis: Why Look At Nemotron 3.5 Super VL’s 25-Cent Task Price? on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get hardware and tech essentials 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 on October 15, with Artificial Analysis reporting an Intelligence Index score of 26 and a cost of about $0.25 per benchmark task. The figures make the model a fast, comparatively inexpensive option for some deployments, but they do not make it the strongest model for the money: several open-weight alternatives score higher at similar or lower benchmark costs.

Artificial Analysis lists the model at 26.5 points and places it first among 67 open-weight models in its 40-billion-to-150-billion-parameter size class. The model has 120 billion total parameters, with 12 billion active in its mixture-of-experts design, and a reported one-million-token context window. That size-class ranking is narrower than a claim that it leads all open-weight models: the source says it ranks about 18th when models across sizes are counted.

The reported API rates are $0.20 per million input tokens and $0.575 per million output tokens, with cached input at $0.055 per million. Artificial Analysis estimates one Intelligence Index task costs $0.25. That is a benchmark estimate, not a fixed charge for every customer request. The source says about 89% of that task cost comes from input, including cache writes, while output accounts for less than three cents.

Speed depends heavily on where the model is served. A pre-release DeepInfra endpoint produced about 630 output tokens per second on 100,000-token prompts, according to the source. Across six providers, Artificial Analysis lists a median of 307 tokens per second at that prompt length, with a reported range of 106 to 534. Median time to the first token is 0.89 seconds, while the source estimates roughly 11 seconds of reasoning before the first answer token and 13.6 seconds for a 500-token response.

What the Quarter-Dollar Benchmark Shows

The $0.25 task estimate gives teams a common benchmark comparison, but actual operating costs depend on the prompts, caching and provider used. A workflow that repeatedly supplies a large repository or document collection may be dominated by input charges, even when its output is short. Teams considering deployment should model their own token patterns rather than treating a benchmark average as a quote.

Price is only one part of the decision. According to the source’s comparison, Xiaomi’s MiMo-V2.6-Pro scores 46 at $0.13 per task, while MiMo-V2.6-Flash scores 38 at $0.06. DeepSeek V4.1 Flash is listed at 39 for $0.27, and Inkling Small at 26 for $0.09. Those figures put pressure on a value argument based on raw benchmark intelligence alone.

Nemotron’s case instead includes speed, licensing and self-hosting. The source describes its OpenMDW-1.1 license as permitting commercial use and says the weights are available in NVFP4. For organizations that need a permissive license or want to run a model on their own NVIDIA hardware, those characteristics may outweigh a lower score elsewhere. They do not, by themselves, establish lower total costs or better performance in a specific application.

How the Release Compares With Its Predecessor

The release follows Nemotron 3 Super. Artificial Analysis scores the predecessor at 13, compared with 26.5 for Super VL. The source also reports that the estimated cost per Intelligence Index task fell from $1.64 to $0.25. It attributes the lower task cost partly to fewer generated output tokens: about 46,000 per index task for Super VL versus 114,000 for Nemotron 3 Super in the launch write-up.

The source reports improvements in particular evaluation areas, not just the overall index. On GDPval-AA, a test of economically useful work, Super VL scored 1,162 Elo, up from 490 for its predecessor. Its AutomationBench-AA score was 14.8%, compared with 3.0% for Nemotron 3 Ultra, while the source cautions that the new score remains low in absolute terms. The reported AA-Omniscience hallucination rate fell from 87% to 36% relative to Nemotron 3 Super.

The licensing comparison also needs a boundary. The source calls Super VL the top-scoring permissively licensed US open-weight model, but says Meta’s Muse Spark Open scored 39 and uses a license with commercial-use restrictions. The ranking depends on how models are grouped and which licensing terms a user accepts; it is not a universal ranking of every model for every purpose.

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

— Thorsten Meyer, ThorstenMeyerAI.com

Where Benchmark Results Leave Questions

The figures supplied do not establish how Super VL will perform on a particular company’s tasks. Intelligence Index results and the $0.25 task estimate describe a benchmark setting; production workloads can differ in prompt length, caching, output volume and provider pricing. The source does not provide a workload-specific cost comparison or independent measurements across customers.

Speed results are also not interchangeable. The 630-token-per-second figure came from one pre-release DeepInfra endpoint, while the reported multi-provider median at the same long prompt length was 307 tokens per second. The available material does not identify the eventual speed each customer will receive or explain all differences among providers.

The supplied material ends while discussing evaluation results and does not provide a full account of the model’s image performance, task-level failures or safety testing. It also does not establish the complete costs of self-hosting, including hardware and operations. Those details matter before teams treat the benchmark price or license as a complete deployment assessment.

What Teams Should Test Before Deployment

The next practical step for prospective users is to test the model on the endpoint and workload they expect to run. That means measuring latency and throughput with their usual prompt lengths, checking how caching changes input charges, and comparing task quality against alternatives under the same conditions. The reported provider spread makes a single headline speed figure a poor substitute for such testing.

Teams should also review the applicable license and verify whether its terms fit their intended use. The supplied source reports commercial use is allowed under OpenMDW-1.1, but the article’s benchmark comparisons do not answer legal or operational questions for individual deployments. No further product milestone or update is specified in the supplied material; performance, pricing and provider availability may need to be checked again as deployments develop.

Key Questions

What does the $0.25 task price represent?

It is Artificial Analysis’s estimated cost for one Intelligence Index task, as reported by ThorstenMeyerAI.com. It is a benchmark-based estimate, not a guaranteed price for an individual request or production workload.

Is Nemotron 3.5 Super VL the top-scoring open-weight model?

Not across all models, according to the supplied comparison. It ranks first among 67 open-weight models in the 40-billion-to-150-billion-parameter class, but about 18th when open-weight models of all sizes are counted.

How fast is it?

A pre-release DeepInfra endpoint reached about 630 output tokens per second on 100,000-token prompts. Across six providers, the reported median at that prompt length was 307 tokens per second, with results varying by provider.

What is the case for using it if competitors score higher?

The source points to speed, a permissive commercial-use license and the option to run the model on NVIDIA hardware. Whether those advantages matter more than benchmark score or cost depends on the user’s workload, infrastructure and licensing needs.

Source: ThorstenMeyerAI.com