🔍 Read the full analysis: Nemotron 3.5 And The Open-Weight Gap: A Score Boost With Limits on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
NVIDIA released Nemotron 3.5 Super VL and Bolt 74B-A9B on October 15, 2026. Artificial Analysis rated Super VL at 26, double its predecessor’s score, while its rankings still put several Chinese and UAE models ahead; Meta’s higher-scoring Muse Spark Open carries commercial restrictions.
The two releases are Nemotron 3.5 Super VL and Nemotron 3.5 Bolt 74B-A9B. Both accept text and images and return text, and both use a mixture-of-experts design, which activates only part of the model for each token. Super VL has 120 billion total parameters, with 12 billion active, and a one-million-token context window. Bolt has 74 billion total parameters, with nine billion active, and a 256,000-token context window.
Artificial Analysis scored Super VL at 26 (26.5 unrounded) and Bolt at 21 (21.4 unrounded). Super VL’s predecessor, Nemotron 3 Super, scored 13 at the same stated model size. The evaluation also found that Super VL used about 49,000 output tokens per index task, compared with about 114,000 for its predecessor, and put the cost per task at $0.25 versus $1.64. These are Artificial Analysis’s reported measurements; service speed and price can vary by provider.
Super VL uses NVIDIA’s NVFP4 format, a four-bit representation intended to reduce memory requirements. Artificial Analysis reported that its NVFP4 version scored within 0.1 points of the full-precision BF16 version on the overall index, with no individual evaluation differing by more than 2.6 points. The model is available under the OpenMDW-1.1 license, which the source says permits commercial use. Bolt did not yet have a public API or a listed license in Artificial Analysis’s model information at publication.
Infographic 01 · Four Nemotron models, seven months
Twice the score, less than half the tokens, a seventh of the price.
Nemotron 3.5 Super VL keeps the size of Nemotron 3 Super and doubles its score. Bolt gets within 1.5 points of last summer’s 550B flagship with 9B active parameters.
Intelligence Index
Higher is better. Ten evaluations.
- Nemotron 3 Super12.8
- Nemotron 3 Ultra22.9
- Nemotron 3.5 Super VL26.5
- Nemotron 3.5 Bolt21.4
Model size
Total / active parameters. Bar = total.
- Nemotron 3 Super120.6B / 12.7B
- Nemotron 3 Ultra550B / 55B
- Nemotron 3.5 Super VL120B / 12B
- Nemotron 3.5 Bolt74B / 9B
Output tokens per index task
Lower is better. Thousands of tokens.
- Nemotron 3 Super114k
- Nemotron 3 Ultra28k
- Nemotron 3.5 Super VL49k
- Nemotron 3.5 Bolt82k
Cost per index task
Lower is better. USD, median API price.
- Nemotron 3 Super$1.64
- Nemotron 3 Ultra$0.60
- Nemotron 3.5 Super VL$0.25
- Nemotron 3.5 Boltno API, not measuredn/a
Bolt has no public API yet. It can only be self-hosted, so Artificial Analysis lists no price, speed or cost per task for it. Its 82k output tokens per task is the number to watch.
Source: Artificial Analysis, Nemotron 3.5 Super and Bolt (15 Oct 2026) and model data read 16 Oct 2026. Intelligence Index v4.3.2, ten evaluations. Nemotron 3.5 Super VL output tokens per task: 49k on the model page, 46k in the launch article.
Powered by Thorsten Meyer AI
NVIDIA Gains Ground in Efficient Models
Super VL’s score and measured task cost suggest NVIDIA has improved the performance and efficiency of its open-weight offering. It also scored above the company’s larger Nemotron 3 Ultra, which Artificial Analysis rated at 23, despite Ultra having 550 billion total parameters and 55 billion active. For developers choosing models to run or access through providers, the reported score, token use and license can all affect the practical choice.
The release does not put US models at the top of the broader rankings. Artificial Analysis lists models from Chinese labs with scores from 44 to 46, and a model from the UAE’s Institute of Foundation Models at 31. The ranking depends in part on what counts as open weight and on license terms: the source distinguishes models with permissive licenses from those whose licenses impose commercial conditions.
NVIDIA Nemotron 3.5 Super VL AI model
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Rankings Compare
Artificial Analysis published its evaluation on the day NVIDIA announced the models. Its Intelligence Index aggregates results across evaluations; the figures cited here are the scores and comparisons reported in the source material. Super VL also scored above Meta’s Muse Glimmer, which the source lists at 17, and Thinking Machines’ Inkling Small at 25.7.
Meta released Muse Spark Open on October 15 as well. Artificial Analysis scored it at 39, above Super VL’s 26, and reported its cost per index task as $0.84 compared with Super VL’s $0.25. Muse Spark Open is much larger, at 816 billion total parameters and 42 billion active. The source says it uses the Muse Model License, which includes commercial restrictions, while Super VL’s OpenMDW-1.1 permits commercial use. That distinction means a higher score does not by itself settle which model is suitable for a given user.
In the rankings described by the source, Super VL placed 18th among listed open-weight models and 11th among those with permissive licenses. The source says nine of the ten highest-scoring open-weight models came from Chinese labs, counting each model once. Those standings describe the Artificial Analysis listing at the time of publication and may change as models or evaluations are added.
““NVIDIA’s own progress is excellent. America’s open-weight position has improved, but it has not changed.””
— Thorsten Meyer, ThorstenMeyerAI.com
License and Deployment Details Pending
Artificial Analysis had not listed a license for Bolt or a public API for the model at publication, so its commercial terms and API pricing were not established in the source material. Super VL’s reported $0.25 task cost is an evaluation estimate, not a universal price for every deployment.
Speed figures also vary by provider. Artificial Analysis measured about 630 output tokens per second on a pre-release DeepInfra endpoint using NVFP4 weights with 100,000-token prompts. The source reports a median of 196 tokens per second across six providers listed on the model page. It is not yet clear how broadly the pre-release result will translate to other hosting setups.
Provider Access and Rankings Ahead
Further provider listings and published license information will clarify how Bolt can be used and what it costs to run through hosted services. For Super VL, results across more providers will show how closely real-world speed and pricing match the measurements reported on release day.
Artificial Analysis’s rankings may also shift as it evaluates more models. The current comparison establishes a substantial gain for NVIDIA’s latest model over its predecessor, while the relative position of US models remains subject to new releases, licensing terms and evaluation results.
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 accept text and images and return text.
How did Super VL score?
Artificial Analysis rated Super VL at 26 on its Intelligence Index, or 26.5 before rounding. Its predecessor, Nemotron 3 Super, scored 13 in the comparison reported by the source.
Does Super VL lead all open-weight models?
No. It led the permissively licensed US models in the source’s account, but Artificial Analysis listed higher-scoring models from China and the UAE. Meta’s Muse Spark Open also scored 39, though the source says its license has commercial restrictions.
What is known about Bolt’s price and license?
The source says Bolt had no public API and that Artificial Analysis had not yet listed its license. Its hosted API price and task cost were therefore not established at publication.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
