Could Nemotron 3.5 Super VL’s 26 Score Fit Your AI Needs?

🔍 Read the full analysis: Could Nemotron 3.5 Super VL’s 26 Score Fit Your AI Needs? on ThorstenMeyerAI.com

NVIDIA released Nemotron 3.5 Super VL on October 15, and Artificial Analysis scored it 26 on its Intelligence Index. The model pairs that benchmark result with reported high throughput, low evaluation-task costs and a license that allows commercial use, but the score alone does not make it a leading value choice: several alternatives in the source comparison score higher for similar or less money.

Artificial Analysis reports a score of 26.5, rounded to 26, and ranks Super VL first among 67 open-weight models in its 40-billion-to-150-billion-parameter size class. It ranks about 18th among open-weight models overall, according to the source. The model has 120 billion total parameters and 12 billion active parameters in a mixture-of-experts design, accepts text and images, and has a one-million-token context window. NVIDIA released its weights in NVFP4, a four-bit format.

The reported API prices are $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. The source compares that with $1.64 for Nemotron 3 Super, its predecessor, and says the newer model scored twice as high. Those figures describe the benchmark evaluation and listed service pricing; actual costs depend on how a team uses the model and which provider it selects.

Speed figures vary by measurement. A pre-release DeepInfra endpoint serving NVFP4 weights produced about 630 output tokens per second on 100,000-token prompts, according to the source’s account of Artificial Analysis data. 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. The reported median time to first token is 0.89 seconds, while reasoning can delay the first answer token to about 11 seconds.

Where Speed and Cost Matter

Super VL’s strongest case may be operational rather than a top raw benchmark score. Its commercial-use license, open weights and option to run the model on NVIDIA hardware may matter to organizations that need control over deployment or want to keep data within their own infrastructure. The source identifies its license as OpenMDW-1.1 and says it allows commercial use.

Cost comparisons depend on the task mix. The source says roughly 89% of the $0.25 evaluation-task cost comes from input, with cache writes accounting for 69% and output costing less than three cents. Teams that repeatedly send large documents or code repositories may therefore see their bills shaped more by input and caching than by output-token prices. The benchmark cost is a useful reference point, not a forecast for every workload.

For interactive products, throughput is only part of the experience. The reported 0.89-second median time to first token is relatively quick, but the source says reasoning takes about 11 seconds before the first answer token and a 500-token answer takes about 13.6 seconds end to end. That profile may suit background processing better than a chat experience that needs an immediate response.

Benchmarks Behind the Release

The Intelligence Index provides an aggregate score, but the source also points to substantial differences across evaluation areas. It reports that GDPval-AA reached 1,162 Elo, up from 490 for Nemotron 3 Super, and above the cited results for Nemotron 3 Ultra and Inkling Small. On AA-Briefcase, Super VL scored 999 Elo versus 876 for Nemotron 3 Ultra, according to the source.

On SaaS automation, the source reports a 14.8% result on AutomationBench-AA, compared with 3.0% for Nemotron 3 Ultra. It characterizes that as a large improvement while also noting the absolute result remains low. On AA-Omniscience, it reports a hallucination rate of 36%, down from 87% for the predecessor. These are results on named tests, not guarantees about accuracy or automation performance in a particular deployment.

Other models in the source comparison scored higher on the index at similar or lower evaluation-task costs. Xiaomi’s MiMo-V2.6-Pro scored 46 at $0.13 per task, while its MiMo-V2.6-Flash scored 38 at $0.06. DeepSeek V4.1 Flash scored 39 at $0.27 and GLM-5.3-Flash scored 42 at $0.28. Inkling Small matched Super VL’s score of 26 at $0.09. The source also cites proprietary tiers Claude Haiku 5.5 at 43 for $0.21 and GPT-6 Luna at 38 for $0.07.

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

— Thorsten Meyer, ThorstenMeyerAI.com

Provider Results and Real-World Fit

The reported 630-token-per-second result came from one endpoint, while the median across six providers was 196 tokens per second on standard prompts. The source gives a 100,000-token-prompt median of 307 and a range from 106 at the fifth percentile to 534 at the 95th percentile. Results may differ at the endpoint a customer actually uses.

The source does not establish how Super VL will perform on each organization’s own data, software or latency requirements. Benchmark scores and estimated task costs cannot settle that question. The model’s ranking among permissively licensed U.S. open-weight models also depends on the licensing criteria: the source says Meta’s Muse Spark Open scored 39 but has commercial-use restrictions under its Muse Model License. Teams should check the license terms that apply to their use.

Testing Before Deployment

The next practical step for teams considering Super VL is to benchmark it on their own tasks and through the provider or hardware they plan to use. That should include the full response time, accuracy on representative prompts, image handling where relevant, and input and cache costs—not only peak output throughput or the Intelligence Index task estimate.

Provider measurements and real-world deployments may add evidence about how the model performs at different prompt lengths and workloads. The source does not give a schedule for further benchmarks or a specific next release milestone, so the timing of those updates is unclear.

Key Questions

What does Nemotron 3.5 Super VL’s score of 26 mean?

It is the model’s reported result on Artificial Analysis’ Intelligence Index. According to the source, it ranks first among 67 open-weight models in its 40-billion-to-150-billion-parameter size class and about 18th among open-weight models overall.

Is Super VL the fastest model available?

The source reports about 630 output tokens per second on a pre-release DeepInfra endpoint with 100,000-token prompts. Across six providers, the reported median was 307 tokens per second for that prompt length. The evidence supports a strong speed result on one endpoint, not a universal fastest-model claim.

How much does it cost to use?

The listed API prices 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 at $0.25. Actual costs depend on the prompts, caching and provider.

Can businesses use the model commercially?

The source identifies the license as OpenMDW-1.1 and says it allows commercial use. Organizations should review the license terms against their intended deployment before relying on that assessment.

Does the score make Super VL the best-value model?

Not by the source’s benchmark score and estimated task-cost comparison alone. It lists several models with higher scores at similar or lower estimated costs. Super VL may still suit teams that place extra weight on speed, its stated license, NVIDIA hardware or self-hosting.

Source: ThorstenMeyerAI.com