Nemotron 3.5 Super VL At A Glance: Cost, Latency, And Score

🔍 Read the full analysis: Nemotron 3.5 Super VL At A Glance: Cost, Latency, And Score on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets 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, 2026, and Artificial Analysis reported an Intelligence Index score of 26 that day, double the 13 recorded for its predecessor. The model combines a comparatively low reported cost of $0.25 per index task with high speed on one pre-release endpoint, but rival models in the supplied comparison score higher for similar or lower task costs.

Artificial Analysis lists the model at 26.5 points and ranks it first among 67 open-weight models in its 40-billion-to-150-billion-parameter size class. That is a size-specific ranking, not a claim that it leads all open-weight models: the source places it around 18th when models across sizes are counted. Nemotron 3.5 Super VL has 120 billion total parameters, with 12 billion active in its mixture-of-experts design, and accepts text and images as input while producing text.

The reported API rates are $0.20 per million input tokens and $0.575 per million output tokens; cached input is listed at $0.055 per million tokens. Artificial Analysis puts the cost of an index task at $0.25. The source says the predecessor cost $1.64 per task, but that comparison reflects the reported evaluation task and should not be treated as a universal production bill. The weights are released in NVFP4, a four-bit format, and six providers are listed.

Speed figures depend on the measurement. A pre-release DeepInfra endpoint handling 100,000-token prompts produced about 630 output tokens per second, which Artificial Analysis described as its fastest measured result at that prompt length for a model scoring at least 20. Across six providers, the model page reports median output speeds of 196 tokens per second on its standard prompt and 307 with 100,000-token prompts. The reported median time to first token is 0.89 seconds, while 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.

Cost and Speed Versus Score

The results matter most to teams choosing between raw benchmark performance and operational considerations such as throughput, control over deployment and licensing. The source’s comparison lists Xiaomi’s MiMo-V2.6-Pro at 46 points for $0.13 per index task, MiMo-V2.6-Flash at 38 for $0.06, and Inkling Small at 26 for $0.09. DeepSeek V4.1 Flash and GLM-5.3-Flash are listed at 39 and 42 points, respectively, for $0.27 and $0.28 per task. These are comparisons from the supplied source, not guarantees of equal performance on a reader’s own workload.

That makes Super VL’s strongest case less about leading on score per dollar and more about its combination of speed, an open-weight release and a license the source says allows commercial use. The source identifies the license as OpenMDW-1.1. Teams that need to host weights on their own NVIDIA hardware may value that option, while others may prefer a higher-scoring or cheaper model available through their chosen provider. The reported 11-second reasoning period also means high token throughput does not automatically translate into a quick conversational response.

Costs may also vary with how an application uses the model. The source estimates that about 89% of the $0.25 task cost comes from input, including 69% from cache writes, while output costs less than three cents. Those figures describe the index-task breakdown. They suggest that organizations sending large documents repeatedly should test their own input and caching patterns rather than infer total operating cost from the output-token price alone.

How the Index Results Compare

Nemotron 3.5 Super VL follows NVIDIA’s Nemotron 3 Super. The source reports a rise from 13 to 26 points on Artificial Analysis’s Intelligence Index, alongside a decrease in output tokens used during an index task from about 114,000 to 46,000. It also reports a reduction in the model’s AA-Omniscience hallucination rate from 87% to 36%. These are benchmark results reported in the supplied material; they do not establish a hallucination rate for every real-world use.

The source describes gains on particular evaluations: GDPval-AA, covering economically useful work, is listed at 1,162 Elo, compared with 490 for Nemotron 3 Super; AA-Briefcase is listed at 999 Elo, against 876 for Nemotron 3 Ultra. AutomationBench-AA is reported at 14.8%, compared with 3.0% for Ultra. These task-specific results help explain why a single aggregate score may not predict performance on a particular workflow.

License comparisons also need qualification. The source says Nemotron 3.5 Super VL is the leading permissively licensed US open-weight model, while Meta’s Muse Spark Open scored 39 but uses a license with commercial restrictions. That distinction may matter to legal and procurement teams. The ranking depends on the source’s definitions of permissive licensing, US origin and the models included.

“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 pre-release endpoint. The source gives a six-provider median and a range of 106 to 534 tokens per second at the 5th and 95th percentiles, but it does not identify which provider a particular customer will use or what performance that customer will receive. Actual speed and latency remain dependent on endpoint, hardware, prompt length and workload.

The supplied material does not provide details on how the index score maps to any specific business task, nor does it report independent production trials, reliability over time or the complete methodology behind each comparison. It also does not establish whether the quoted API prices will remain unchanged. The model’s image-input capability is specified, but no image-specific benchmark results are included here. Buyers should treat the benchmark and cost figures as reported measurements and verify them against their own evaluation criteria.

Test the Intended Deployment

Teams considering Nemotron 3.5 Super VL can compare it with alternatives on their own prompts, documents and latency requirements, then measure total cost across input, cache and output tokens. They should also confirm the applicable license terms and whether their intended hosting setup supports the released weights. The source does not announce a future benchmark milestone or a scheduled change to pricing, so the next meaningful evidence is likely to come from provider-specific measurements and independent evaluations as deployments proceed.

Key Questions

What score did Nemotron 3.5 Super VL receive?

Artificial Analysis reported an Intelligence Index score of 26; the source lists the more precise figure as 26.5. It ranks first among 67 open-weight models in the cited 40-billion-to-150-billion-parameter class, but around 18th among open-weight models overall.

How much does one index task cost?

The reported cost is $0.25 per Artificial Analysis index task. That is an evaluation figure, not a fixed price for every production request; actual bills depend on token use, caching and the provider’s rates.

How fast is the model?

A pre-release DeepInfra endpoint reached about 630 output tokens per second on 100,000-token prompts. The reported median across six providers was 307 tokens per second for that prompt length, with substantial variation among providers.

Does the model have a permissive commercial license?

The supplied source identifies its license as OpenMDW-1.1 and says commercial use is allowed. Organizations should check the license text and their own legal requirements before deployment.

Is it the best-value open model in the comparison?

Not by the source’s reported score and index-task cost alone. Several listed models score higher for similar or lower costs. Nemotron 3.5 Super VL may still suit teams that prioritize its speed, license, NVIDIA hardware deployment or other operational requirements.

Source: ThorstenMeyerAI.com