Can A Score Of 26 Make Nemotron 3.5 Super VL Worth A Look?

🔍 Read the full analysis: Can A Score Of 26 Make Nemotron 3.5 Super VL Worth A Look? on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get smart everyday buys 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 Oct. 15, and Artificial Analysis rated it 26.5 on its Intelligence Index, roughly double its predecessor’s score of 13. The 120-billion-parameter open-weight model pairs that result with low reported task costs and high throughput, though comparisons in the source show competing models scoring higher for similar or lower cost.

Super VL has 120 billion total parameters, with 12 billion active in its mixture-of-experts design, and a reported context window of one million tokens. It accepts text and images and produces text. Its weights are available in NVFP4, and the source says its OpenMDW-1.1 license permits commercial use. Artificial Analysis lists API prices of $0.20 per million input tokens and $0.575 per million output tokens, with cached input at $0.055 per million tokens.

Artificial Analysis puts the cost of one Intelligence Index task at $0.25, compared with $1.64 for Nemotron 3 Super. The source attributes the lower cost partly to less generated output: about 46,000 output tokens per index task, versus 114,000 for the predecessor. It says roughly 89% of Super VL’s task cost is input, including cache writes, while output costs less than three cents. Those figures describe the index task and do not establish what a particular team’s workloads will cost.

Speed depends on the serving endpoint. A pre-release DeepInfra endpoint running NVFP4 weights reportedly generated about 630 output tokens per second with prompts of 100,000 tokens. Across six providers, Artificial Analysis lists a median of 196 tokens per second on its standard prompt and 307 with 100,000-token prompts; the reported range is 106 to 534 tokens per second. Median time to first token is 0.89 seconds, while the source says reasoning can take about 11 seconds before the first answer token.

Cost and Speed Trade Off Against Score

The score matters because it puts the model’s reported efficiency in perspective. The source lists MiMo-V2.6-Pro at 46 points for $0.13 per task, MiMo-V2.6-Flash at 38 for $0.06, DeepSeek V4.1 Flash at 39 for $0.27, and GLM-5.3-Flash at 42 for $0.28. Inkling Small is listed at 26 for $0.09. These comparisons suggest Super VL’s case is not that it delivers the most Intelligence Index points for the money.

Its appeal may instead be its reported throughput, low first-token latency, commercial-use license and option to run released weights on NVIDIA hardware. Those features can matter to teams with specific infrastructure or deployment requirements. Whether they outweigh lower scores or alternatives’ lower task costs depends on the application, provider and workload; the source’s figures alone do not settle that choice.

The launch material also reports gains in selected evaluations. GDPval-AA, a benchmark for economically useful tasks, rose to 1,162 Elo from 490 for Nemotron 3 Super. AutomationBench-AA is reported at 14.8%, versus 3.0% for Nemotron 3 Ultra, though the source describes the new result as low in absolute terms. On AA-Omniscience, the reported hallucination rate fell from 87% to 36%. These are benchmark results, not guarantees of performance in production.

How the Score Is Ranked

The Intelligence Index score is one comparison point, not a measure of every use case. Artificial Analysis places Super VL first among 67 open-weight models in its 40-billion-to-150-billion-parameter size class, according to the source. Across open-weight models overall, the article estimates its position at about 18th. The class ranking and the broader placement describe different comparison groups.

The source also distinguishes permissively licensed U.S. models from all open-weight options. It says Super VL has the top score among permissively licensed U.S. open-weight models. Meta’s Muse Spark Open, released the same day and rated at 39, uses a license that the source says allows commercial use with restrictions. A team’s legal and deployment requirements can therefore change which comparison is relevant.

Super VL follows NVIDIA’s Nemotron 3 Super, which scored 13 on the same index, according to the article. The reported new score is given as 26 in the headline and 26.5 in the detailed specification. The source’s task-cost, speed and benchmark figures were published alongside the release on Oct. 15, 2026.

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

— Thorsten Meyer, ThorstenMeyerAI.com

Provider Results and Real Workloads

The source does not show how Super VL performs across every task, nor whether benchmark gains will carry over to a particular production workload. It also does not specify the evaluation conditions behind every comparison in enough detail to independently reconcile differences across models. The 630-token-per-second result came from one pre-release endpoint; actual speed will depend on the selected provider and prompt. The reported 106-to-534 range across providers underscores that variation.

Cost will also vary with input length, caching, generated output and usage patterns. The quoted $0.25 figure is for one Intelligence Index task, not a general price for a customer request. The source gives no workload-specific estimates, and it does not establish how performance or costs may change as serving conditions evolve.

Test the Intended Deployment

The practical next step for teams considering Super VL is to evaluate it on their own tasks and measure the endpoint they intend to use. That means checking output quality, latency and token use against alternatives, with the team’s actual prompts and cache behavior. The source recommends measuring the chosen provider because speed varies substantially across the listed services.

No later milestone or additional release schedule is identified in the source. Further provider measurements and task-specific evaluations would help clarify whether Super VL’s speed, license and deployment options compensate for its lower Intelligence Index score relative to several competing models.

Key Questions

What score did Nemotron 3.5 Super VL receive?

Artificial Analysis rated it 26.5 on its Intelligence Index. The source compares that with 13 for Nemotron 3 Super.

How much does an Intelligence Index task cost?

The source lists a cost of $0.25 per index task. That figure is specific to the benchmark task; costs for other workloads depend on token use and caching.

Is Super VL the fastest model in the comparison?

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

Can businesses use the model commercially?

The source says Super VL is released under the OpenMDW-1.1 license, which allows commercial use. Teams should check the license terms against their own requirements.

Does a score of 26 make it the best value?

Not on Intelligence Index points per task cost in the comparisons provided. Several listed open models score higher for similar or lower task costs; Super VL’s potential advantages include throughput, latency, license and the option to run its weights on NVIDIA hardware.

Source: ThorstenMeyerAI.com