A Look At Nemotron 3.5 Super VL’s Score Of 26

🔍 Read the full analysis: A Look At Nemotron 3.5 Super VL’s Score Of 26 on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get everyday 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, and Artificial Analysis scored it 26 on its Intelligence Index, double the score of its predecessor. The model combines low reported task costs and high throughput with a permissive commercial-use license, though benchmark comparisons show that several alternatives deliver higher scores for similar or lower prices.

Artificial Analysis lists the model at 26.5 points in its detailed data and places it first among 67 open-weight models in the 40-billion-to-150-billion-parameter size class. It has 120 billion total parameters, with 12 billion active in its mixture-of-experts design, and accepts text and images. NVIDIA released NVFP4, or 4-bit, weights. The model page lists a one-million-token context window and the OpenMDW-1.1 license, which permits commercial use.

The service pricing listed by Artificial Analysis is $0.20 per million input tokens and $0.575 per million output tokens, with cached input priced at $0.055 per million. One Intelligence Index task costs about $0.25 by the benchmark’s estimate. That is down from $1.64 for Nemotron 3 Super, while the score rose from 13 to 26. Artificial Analysis attributes much of the task-cost difference to lower output volume: about 46,000 output tokens per index task for Super VL, compared with 114,000 for its predecessor.

Speed varies by measurement. A pre-release DeepInfra endpoint running the NVFP4 weights produced about 630 output tokens per second on prompts of 100,000 tokens, according to Artificial Analysis. Across six providers, the model page reports median output speeds of 196 tokens per second on standard prompts and 307 on 100,000-token prompts, with the measured range stretching from 106 to 534 tokens per second. Median time to first token is 0.89 seconds, while the first answer token takes about 11 seconds when the model reasons before responding.

Speed Comes With Tradeoffs

The results make deployment speed and control central parts of the model’s appeal. Teams that need to process long prompts, run background agents or keep workloads on their own NVIDIA hardware may value its throughput, relatively low estimated task cost and commercial-use license. The provider-to-provider speed spread also means that a headline rate from one endpoint may not predict performance in a team’s chosen service.

For buyers comparing models mainly by benchmark score per dollar, the 26-point result is less compelling. Artificial Analysis lists MiMo-V2.6-Pro at 46 points for $0.13 per task, MiMo-V2.6-Flash at 38 for $0.06, and Inkling Small at 26 for $0.09. Those estimates use the index’s task-cost methodology and may not match a particular workload. Super VL’s case therefore depends on factors such as speed, licensing and self-hosting, rather than leading the listed price-to-score comparisons.

The benchmark profile is uneven in useful ways. Artificial Analysis reports improvements in knowledge-work tasks and SaaS automation compared with Nemotron 3 Ultra, alongside a lower measured hallucination rate than Nemotron 3 Super. The automation score remains low in absolute terms, however, and a benchmark average cannot establish how the model will perform in a specific company’s workflow.

How the Score Compares

The 26-point score is a benchmark result, not a direct measure of every deployment’s quality. Artificial Analysis places Super VL at about 18th among open-weight models overall when counting each model once, despite its first-place position in the cited size class. The distinction matters: its class ranking compares models with roughly 40 billion to 150 billion total parameters, while the broader ranking includes models outside that band.

Price comparisons in the source include other open models and lower-cost proprietary offerings. For example, it lists DeepSeek V4.1 Flash at 39 points for $0.27 per task and GLM-5.3-Flash at 42 for $0.28. Claude Haiku 5.5 at maximum effort is listed at 43 for $0.21, while GPT-6 Luna at maximum effort is listed at 38 for $0.07. The source also says Super VL leads among permissively licensed US open-weight models, but that description does not make it the leading model under every licensing or performance comparison.

Provider Results Still Vary

The published speed figures do not guarantee the same results for every deployment. Artificial Analysis reports a 106-to-534-token-per-second range across providers, while the 630-token figure came from one pre-release endpoint. The source does not establish whether production endpoints will sustain that single-provider rate or how results will vary with different hardware and workloads.

The benchmark and cost figures also leave practical questions open. The $0.25 estimate applies to an Intelligence Index task, and actual bills depend on input volume, caching, output length and provider pricing. The available material gives no independent results for a reader’s own documents, tools or automation tasks. The source excerpt also ends partway through its discussion of per-evaluation strengths, so it does not provide a complete account of the model’s weaker benchmark areas.

Test the Intended Deployment

The next useful step for prospective users is to measure the model on the provider and hardware they plan to use. Teams can compare end-to-end response time, output quality, hallucinations and cost on representative tasks, including long prompts that benefit from cached input. The reported 11-second wait for the first answer token may suit background processing while feeling slow in interactive chat.

Further provider measurements and independent workload evaluations would help clarify how broadly the reported speed and cost advantages hold. For now, Artificial Analysis’s release-day results establish a strong case for testing Super VL where throughput, self-hosting or license terms matter; they do not establish that it is the best-value model for every use.

Key Questions

What score did Nemotron 3.5 Super VL receive?

Artificial Analysis reports a headline Intelligence Index score of 26; its detailed listing gives 26.5. The source compares that with 13 for Nemotron 3 Super.

How much does one benchmark task cost?

Artificial Analysis estimates $0.25 per Intelligence Index task. Actual costs depend on a deployment’s input, cache and output usage and its provider’s prices.

Is it the fastest model in the comparison?

One pre-release DeepInfra endpoint reached about 630 output tokens per second on 100,000-token prompts. Across six providers, the reported median for that prompt length was 307 tokens per second, so results depend on the provider.

Can companies use the model commercially?

The source identifies its license as OpenMDW-1.1 and says commercial use is allowed. Teams should review the license terms for their intended use.

Does a score of 26 make it the best value?

Not by the listed benchmark score and estimated task cost alone. The source names several models with higher scores at similar or lower estimated costs; Super VL’s appeal also includes speed, licensing and the option to run it on NVIDIA hardware.

Source: ThorstenMeyerAI.com