Can Nemotron 3.5 Super VL Deliver Fast AI Tasks At Low Cost?

🔍 Read the full analysis: Can Nemotron 3.5 Super VL Deliver Fast AI Tasks At Low Cost? 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, 2026, with Artificial Analysis reporting an Intelligence Index score of 26.5, a $0.25 cost per benchmark task and high output speed on one provider endpoint. The figures make the model a candidate for teams prioritizing throughput, commercial-use licensing or deployment on NVIDIA hardware, but the score is below several alternatives that cost as much or less.

The model has 120 billion total parameters, with 12 billion active, and accepts text and images as input while producing text. Its reported context window is 1 million tokens. NVIDIA released its weights in NVFP4, a four-bit format, and the source says the OpenMDW-1.1 license allows commercial use. Artificial Analysis lists six providers.

Artificial Analysis measured $0.20 per million input tokens and $0.575 per million output tokens, with cached input priced at $0.055 per million tokens. Its estimated cost for one Intelligence Index task is $0.25. The source reports that Nemotron 3 Super cost $1.64 per task and scored 13, compared with 26.5 for the new model; these are benchmark estimates, not a guaranteed cost for every production workload.

Speed varies by measurement. A pre-release DeepInfra endpoint serving NVFP4 weights produced about 630 output tokens per second with 100,000-token prompts. Across six providers, Artificial Analysis lists median output speeds of 196 tokens per second on its standard prompt and 307 with 100,000-token prompts, with results ranging from 106 to 534 tokens per second. Median time to first token is 0.89 seconds, while the source reports about 11 seconds before the first answer token after reasoning.

Where Low Cost Meets Lower Scores

The release highlights a practical distinction for AI buyers: low price per benchmark task is not the same as high capability per dollar. In the comparisons supplied by ThorstenMeyerAI.com, MiMo-V2.6-Pro scored 46 at $0.13 per task, MiMo-V2.6-Flash scored 38 at $0.06, and GLM-5.3-Flash scored 42 at $0.28. Inkling Small scored 26 at $0.09, against Nemotron’s $0.25. These figures are Artificial Analysis comparisons as reported in the source, and performance can differ across real tasks and deployments.

Nemotron’s potential case is instead its speed, commercial-use license and option for self-hosting. Those attributes can matter to organizations with strict data controls, specific hardware investments or licensing needs. The reported spread between providers also means that published peak throughput should not be treated as a deployment guarantee. Teams would need to test the actual provider or hardware configuration against their own workloads and latency targets.

Cost composition may matter as much as headline token prices. The source says input accounted for about 89% of the $0.25 benchmark-task cost, with cache writes alone making up 69%, while output cost was under three cents. Workloads that repeatedly send large documents or code repositories may therefore have bills shaped mainly by input and caching. Those proportions describe the cited benchmark evaluation and may not match every use case.

A Bigger Score Than Its Predecessor

Artificial Analysis gave Nemotron 3.5 Super VL an Intelligence Index score of 26.5, rounded to 26, versus 13 for Nemotron 3 Super. The source says the model ranked first among 67 open-weight models in its size class, defined as 40 billion to 150 billion total parameters. That is a size-class ranking, not a claim that it leads all open-weight models: the source places it at about 18th when all open-weight models are counted.

The reported index-task cost is also lower than the predecessor’s: $0.25 versus $1.64. The source attributes the reduction in part to fewer output tokens per index task, about 46,000 compared with 114,000 for Nemotron 3 Super, as well as lower cached-input pricing. These benchmark figures give a basis for comparing the models, but do not establish the cost or accuracy of a specific customer’s application.

The release’s reported strengths are uneven across evaluations. The source lists a GDPval-AA score of 1,162 Elo, up from 490 for Nemotron 3 Super, and an AA-Briefcase score of 999. AutomationBench-AA is reported at 14.8%, compared with 3.0% for Nemotron 3 Ultra. On AA-Omniscience, the source says the hallucination rate fell from 87% to 36%. These are named benchmark results, not a guarantee that the model will perform similarly in every business setting.

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

— Thorsten Meyer, ThorstenMeyerAI.com

Provider Results and Workload Fit

The 630-token-per-second result came from a single pre-release DeepInfra endpoint; it does not describe typical speed across all providers. Artificial Analysis reports a wide provider range, but the supplied material does not identify the conditions behind every provider’s result or establish how those measurements translate to a particular production setup.

The source also does not provide independent details on the benchmark methodology, confidence ranges or the exact version and operating conditions used for every comparison. It reports that commercial use is allowed under OpenMDW-1.1, but organizations should review the license terms themselves before deployment. Actual costs will depend on prompt length, caching, output volume, hardware and provider pricing.

It remains unclear from the supplied material how the model will perform on individual customer workloads, how quickly providers will offer stable versions, and whether benchmark gains will hold across more tasks. The Intelligence Index score and task cost are useful comparison points, not proof of general superiority or savings.

Test Before Production Deployment

The immediate next step for prospective users is testing the chosen endpoint or self-hosted configuration on representative tasks. Teams can compare response quality, first-token delay, throughput and full input-and-output costs with alternatives, while checking how repeated context affects caching charges. For applications requiring image input, evaluation should include the specific visual tasks the system will handle.

Artificial Analysis’s provider listings and benchmark measurements may change as additional services become available or results are updated. The source does not give a schedule for further NVIDIA releases or independent assessments. Until more deployment evidence is available, the reported figures support a targeted evaluation rather than a blanket conclusion that Nemotron 3.5 Super VL is the best-value model.

Key Questions

What score did Nemotron 3.5 Super VL receive?

Artificial Analysis reported an Intelligence Index score of 26.5, usually rounded to 26. The source says it ranked first among 67 open-weight models in its 40-billion-to-150-billion-parameter size class, not across all open-weight models.

How much does a benchmark task cost?

Artificial Analysis lists an estimated $0.25 per Intelligence Index task. That is a benchmark estimate; costs for a live application depend on its input and output volume, caching, provider and deployment setup.

Is the model consistently fast across providers?

No single speed figure captures all providers. A pre-release DeepInfra endpoint reached about 630 output tokens per second on 100,000-token prompts, while the reported median across six providers was 307 tokens per second on that prompt length. The reported range was 106 to 534.

Can businesses use Nemotron 3.5 Super VL commercially?

The source says the model uses the OpenMDW-1.1 license, which allows commercial use. Businesses should review the license terms and confirm that their planned use meets their legal and operational requirements.

Does the score make it the best-value model?

Not on the comparisons in the supplied source. Several competing models scored higher at a similar or lower estimated task cost. Nemotron’s reported advantages instead include speed on some endpoints, commercial-use licensing and the option to run it on NVIDIA hardware.

Source: ThorstenMeyerAI.com