Does Nemotron 3.5’S Score Shift The Open-Weight Landscape?

🔍 Read the full analysis: Does Nemotron 3.5’S Score Shift The Open-Weight Landscape? 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 and Nemotron 3.5 Bolt 74B-A9B on October 15, and Artificial Analysis scored them at 26 and 21 on its Intelligence Index. Super VL’s score is double that of its predecessor, but the current rankings still put leading Chinese and UAE open-weight models ahead, leaving the effect on the wider competition limited.

Artificial Analysis reported a score of 26.5 before rounding for Super VL, compared with 13 for Nemotron 3 Super. Both models are mixture-of-experts reasoning systems that accept text and images and return text. Super VL has 120 billion total parameters, with 12 billion active for each token, and a one-million-token context window. Bolt has 74 billion total parameters, nine billion active, and a 256,000-token context window.

The evaluation also found a change in resource use for Super VL. Its output fell from about 114,000 tokens to 49,000 per index task, while the reported cost per task fell from $1.64 to $0.25. Artificial Analysis measured an NVFP4 version within 0.1 points of full-precision BF16 weights on the overall index; no individual evaluation differed by more than 2.6 points. These are benchmark and provider measurements, not guarantees of performance or cost for every deployment.

Bolt scored 21, 1.5 points below NVIDIA’s Nemotron 3 Ultra, despite having about one-sixth of Ultra’s active parameters. It has no public API listed in the source material, so a comparable hosted price is unavailable. The reported figures come from Artificial Analysis’s evaluation and model listings, as described in the October 16 source report.

A Stronger Midrange US Option

Super VL gives developers a higher-scoring US model under a license that the source report says allows commercial use. Its reported $0.25 cost per index task, compared with $1.64 for its predecessor, could matter to teams weighing model quality against inference expense. The NVFP4 result may also interest operators running models on their own hardware, since four-bit weights require less memory than 16-bit weights.

Those practical advantages do not settle which model is best for a particular workload. Index scores summarize a set of evaluations, and actual results can vary by task, serving provider, hardware and prompt. Bolt’s long output is a further caveat: it generated about 82,000 output tokens per index task, nearly three times Ultra’s 28,000. Without a public API price, its cost per answer cannot yet be compared on the same basis.

The rankings also put the US progress in perspective. Super VL is 18th among the open-weight models listed by Artificial Analysis and 11th among permissively licensed models, according to the source report. The gain improves NVIDIA’s position among US models in that licensing category, but the reported table does not show it overtaking the leading international models.

How the Rankings Compare

Artificial Analysis published its evaluation on the same day NVIDIA released the models. The source report says China’s MiMo-V2.6-Pro scored 46, while GLM-5.3, Step 5 and Kimi K3 scored between 44 and 45. Those models carry licenses with commercial conditions, according to the report. The UAE’s K2 Horizon 375B A23B scored 31 and is listed as Apache 2.0 licensed.

Meta released Muse Spark Open on October 15 as well. Artificial Analysis scored it at 39, with a reported cost of $0.84 per index task. It is substantially larger than Super VL: 816 billion total parameters and 42 billion active. The source report says Muse Spark Open uses the Muse Model License, which has commercial restrictions, while Super VL uses OpenMDW-1.1, which allows commercial use.

The licensing distinction changes how the table can be read. Super VL is described as the leading US model among permissively licensed models in Artificial Analysis’s comparison. That does not make it the top-scoring open-weight model overall. The source report also says nine of the ten highest-scoring models in the listed open-weight ranking come from Chinese labs, with Meta’s Muse Spark Open in the remaining position.

Limits of the Current Results

The benchmark scores do not show how the models will perform across all production tasks, and provider speed and pricing can differ. For Super VL, Artificial Analysis measured about 630 output tokens per second on a pre-release DeepInfra endpoint with 100,000-token prompts, but the median across six providers listed on the model page was 196 tokens per second. The source report does not establish that other providers will match the pre-release result.

Bolt’s commercial license details and hosted price are not available in the supplied material; Artificial Analysis had not listed its license, and no public API was available. The source material also ends before giving further details about Muse Spark Open’s license terms. Its reported commercial restrictions should therefore be understood at that level, without assuming additional conditions. Rankings may change as more models and evaluations are added.

Deployment and Rankings Ahead

The next useful evidence will come from public deployment details for Bolt, including its license, API availability and pricing. For Super VL, measurements across more providers and real workloads will help show whether its benchmark cost and speed figures carry over beyond the reported evaluation setup.

Artificial Analysis’s model pages can change as evaluations and provider listings are updated. Readers comparing the models should check the evaluation date, license terms, task-specific scores and serving conditions when those details become available. The October 15 release establishes a sizable improvement for NVIDIA’s own models; whether that changes the broader rankings remains an open question.

Key Questions

What did NVIDIA release?

NVIDIA released Nemotron 3.5 Super VL and Nemotron 3.5 Bolt 74B-A9B on October 15, 2026. Both are open-weight reasoning models that accept text and images and return text.

How did Super VL score?

Artificial Analysis gave Super VL an Intelligence Index score of 26, or 26.5 before rounding. Its predecessor, Nemotron 3 Super, scored 13 in the comparison reported by ThorstenMeyerAI.com.

Does Super VL lead all open-weight models?

No. The source report places it 18th among the open-weight models listed by Artificial Analysis and 11th among permissively licensed models. It describes Super VL as the leading US model among permissively licensed models.

What is not yet known about Bolt?

The supplied material says Bolt has no public API yet and that Artificial Analysis had not listed its license. A public price and comparable cost per task are also unavailable.

Source: ThorstenMeyerAI.com