How Nemotron 3.5 Super VL Stacks Up On Cost And Latency

🔍 Read the full analysis: How Nemotron 3.5 Super VL Stacks Up On Cost And Latency on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• 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, and Artificial Analysis reported a score of 26 on its Intelligence Index and a cost of $0.25 per index task. The model pairs that benchmark result with fast reported inference, commercial-use licensing and weights that can be run on NVIDIA hardware, but its speed varies by provider and several alternatives score higher or cost less per task.

Artificial Analysis lists the model at 26.5 points, with 120 billion total parameters and 12 billion active parameters in a mixture-of-experts design. It accepts text and images, has a one-million-token context window, and was released with NVFP4 weights. Its listed API rates are $0.20 per million input tokens and $0.575 per million output tokens; cached input is $0.055 per million tokens. Artificial Analysis puts the cost of one index task at $0.25.

The fastest reported result came from a pre-release DeepInfra endpoint serving the NVFP4 weights: about 630 output tokens per second with 100,000-token prompts. That was the fastest result Artificial Analysis had measured at that prompt length among models scoring at least 20, according to its report. Across six providers, however, the model page lists median output speeds of 196 tokens per second on standard prompts and 307 with 100,000-token prompts. Its reported range runs from 106 to 534 tokens per second.

Artificial Analysis reports a median time to first token of 0.89 seconds. The source says the model spends time reasoning before responding: the first answer token takes about 11 seconds, while a 500-token answer takes about 13.6 seconds end to end. The source also reports that Nemotron 3.5 Super VL uses about 46,000 output tokens per index task, compared with 114,000 for Nemotron 3 Super, and costs about 85% less per task than that predecessor, which scored 13.

Where Speed Meets Task Cost

For teams running agents over large documents or code repositories, input usage can drive the bill more than output. The source’s breakdown attributes about 89% of the $0.25 task cost to input, with cache writes alone accounting for 69%; output costs less than three cents. Workloads that repeatedly read the same material may benefit from the model’s low cached-input rate, though actual costs will depend on token use and provider pricing.

Latency and throughput also have different implications. A 0.89-second median first-token time can help interactive systems begin responding quickly, while the reported 11 seconds before the first answer token may be more noticeable in a chat interface. The figures suggest potential fit for background pipelines, but operators should measure the provider and hardware configuration they plan to use.

At 26 points, Super VL is not the strongest value in the supplied comparisons if the sole measure is benchmark score per task dollar. MiMo-V2.6-Pro is listed at 46 points for $0.13 per task, while Inkling Small matches Super VL’s 26 points at $0.09. Super VL’s reported advantages instead include its permissive commercial-use license and the option to run the released weights on a team’s own NVIDIA hardware.

Amazon

NVIDIA Nemotron 3.5 Super VL AI model

View Latest Price

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

How the Model Ranks

The headline ranking depends on the comparison group. Artificial Analysis places Nemotron 3.5 Super VL first among 67 open-weight models in its 40-billion-to-150-billion-parameter size class. The source says it ranks about 18th when compared with all open-weight models, counting each model once. The class-leading result is therefore specific to models of similar total size, rather than the full open-weight field.

The source compares its $0.25 task cost with $1.64 for Nemotron 3 Super, which scored 13. It also lists other models at similar or lower task costs: MiMo-V2.6-Flash at 38 points and $0.06; DeepSeek V4.1 Flash at 39 and $0.27; and GLM-5.3-Flash at 42 and $0.28. These comparisons are reported from the supplied source material and depend on Artificial Analysis’ index and task-cost methodology.

The source describes the model as the top-scoring permissively licensed US open-weight model. That qualification matters: Meta’s Muse Spark Open scored 39 in the source’s comparison but uses a license that permits commercial use with restrictions. Nemotron 3.5 Super VL is released under OpenMDW-1.1, which the source says allows commercial use.

Provider Results Remain Uneven

The 630-token-per-second result was measured on one pre-release endpoint; it does not establish the speed every user will see. The six-provider median is substantially lower, and the reported 106-to-534 range shows that provider choice can make a large difference. The supplied source does not identify the final performance of every provider or say whether the pre-release endpoint result will persist.

Other important details are also absent from the supplied material, including the hardware and serving conditions behind each provider’s result, the precise methodology behind task-cost estimates, and performance on a broader range of real-world workloads. The benchmark score and cost figures should be read as reported measurements, not guarantees for a particular deployment.

The source also reports improvements in knowledge work, SaaS automation and hallucination rate compared with earlier NVIDIA models. It gives specific benchmark results, including a reduction in the reported hallucination rate from 87% to 36%, but does not provide enough detail here to establish how those results translate to every application or what error types remain.

Measure Before Deployment

Teams considering the model can compare providers using their own prompt lengths, caching patterns and response-size requirements. That testing should include end-to-end response time, not just output tokens per second, because time spent reasoning affects how long users wait for an answer.

For production decisions, operators will also need to assess the released weights, licensing terms and hardware requirements against their own systems. Further provider measurements and independent tests on representative workloads could clarify whether the pre-release throughput result is repeatable and whether the reported index-task cost matches deployment bills.

Key Questions

When was Nemotron 3.5 Super VL released?

NVIDIA released Nemotron 3.5 Super VL on October 15, 2026. Artificial Analysis published figures the same day, according to the supplied source.

How much does an Intelligence Index task cost?

Artificial Analysis lists an estimated cost of $0.25 per task. The supplied source says about 89% of that cost is input, but actual bills depend on usage and provider.

How fast is the model?

A pre-release DeepInfra endpoint reached about 630 output tokens per second with 100,000-token prompts. Across six providers, the reported median was 307 tokens per second for prompts of that length; measure the provider you expect to use.

Does its score make it the best value among open models?

Not by score and task cost alone in the supplied comparisons. MiMo-V2.6-Pro scored 46 for $0.13 per task, while Nemotron 3.5 Super VL scored 26 for $0.25. Its reported strengths include speed, a commercial-use license and the option to self-host.

Source: ThorstenMeyerAI.com

×HALLOWEEN

Halloween Picks

halloween

halloween

View Latest Price

As an affiliate, we earn on qualifying purchases.