🔍 Read the full analysis: Nemotron 3.5’S Score Gain Meets A Persistent Open-Weight Gap on ThorstenMeyerAI.com
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
TL;DR
NVIDIA released Nemotron 3.5 Super VL and Bolt on October 15, 2026. Artificial Analysis rated Super VL at 26, up from 13 for its predecessor, but ranked it 18th among listed open-weight models as Chinese models and a UAE model scored higher.
NVIDIA released Nemotron 3.5 Super VL and Nemotron 3.5 Bolt 74B-A9B on October 15, with Artificial Analysis scoring them at 26 and 21 on its Intelligence Index. Super VL doubled the score of its predecessor, but the evaluation placed it 18th among listed open-weight models, behind leading models from China and the United Arab Emirates.
Super VL is the larger model, with 120 billion total parameters and 12 billion active for each token. It accepts text and images, has a reported one-million-token context window and uses the OpenMDW-1.1 license, which allows commercial use. Artificial Analysis reported an unrounded Intelligence Index score of 26.5, shown as 26, compared with 13 for Nemotron 3 Super, a model in the same size class.
The evaluation also reported lower resource use for Super VL on its index tasks: about 49,000 output tokens and $0.25 per task, compared with about 114,000 tokens and $1.64 for the predecessor. Artificial Analysis said the model’s NVFP4 version scored within 0.1 points of its full-precision BF16 version; no individual evaluation differed by more than 2.6 points. Those results describe the tested setup and tasks, not a guarantee for every deployment.
Bolt has 74 billion total parameters, nine billion active, and a 256,000-token context window. It scored 21, close to NVIDIA’s Nemotron 3 Ultra score of 23 despite Ultra’s 550 billion total and 55 billion active parameters. Artificial Analysis ranked Bolt fourth among models below 100 billion total parameters and first among US models in that category. Bolt has no public API listed in the source material, so its serving price could not be compared.
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
A Stronger Model, Same Global Gap
For developers choosing a commercially usable open-weight model, Super VL combines a stronger benchmark result with a permissive license and lower reported cost per index task than its predecessor. Its NVFP4 result may also make local deployment more practical on systems with less memory than full 16-bit weights require. Actual performance and cost will still depend on the hardware, serving provider and workload.
The ranking puts the improvement in perspective. Artificial Analysis listed Super VL in 18th place overall among open-weight models, or 11th when counting only permissively licensed models. The source reports that nine of the ten highest-scoring models came from Chinese labs, while a UAE model also scored above Super VL. NVIDIA has improved its position, but the reported results do not show it leading the overall field.
Products Worth Considering
Jackery Explorer 300 Portable Power Station with SolarSaga 100W Solar Panel
Jackery Solar Generator 1000 V2 with 100W Solar Panel, 1024Wh LiFePO4 Battery
Bose QuietComfort Earbuds (2nd Gen) – True Wireless Bluetooth Earbuds with Active Noise Cancellation
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 combines results across evaluations; the scores cited here are the reported index results, not direct measures of every use case. Super VL’s predecessor, Nemotron 3 Super, scored 13 seven months earlier. NVIDIA’s Nemotron 3 Ultra scored 23, while Super VL scored higher despite having fewer total and active parameters.
The same day, Meta released Muse Spark Open. Artificial Analysis scored it at 39, 13 points above Super VL, and reported a cost of $0.84 per index task versus Super VL’s $0.25. Muse Spark Open has 816 billion total parameters and 42 billion active. Its Muse Model License carries commercial restrictions, according to the source, while Super VL’s OpenMDW-1.1 permits commercial use. License terms therefore affect which models are suitable for particular users, independently of benchmark rank.
The source also cites MiMo-V2.6-Pro from Xiaomi at 46, and GLM-5.3, Step 5 and Kimi K3 at scores between 44 and 45. It describes several of those licenses as carrying commercial conditions. The UAE’s K2 Horizon 375B A23B scored 31 and is listed under Apache 2.0. These comparisons are a snapshot of the models Artificial Analysis listed when it published its evaluation.
Limits of the Published Results
The source gives a dated snapshot, but does not establish how the rankings may change as more models are evaluated. It also does not provide a public API price for Bolt or a cost-per-task measurement for it, leaving its operating cost unclear. Speed figures vary by provider: Artificial Analysis measured about 630 output tokens per second on a pre-release DeepInfra endpoint using NVFP4 weights and 100,000-token prompts, while the median across six providers listed on the model page was 196 tokens per second.
Benchmark scores and measured costs do not show how either model will perform on every task or hardware setup. The source material also ends partway through its discussion of Muse Spark Open’s license, so it does not establish the full scope of that license’s restrictions.
Provider Access and Licensing
The next practical comparison depends on broader provider availability and published pricing, particularly for Bolt, which the source says had no public API at release. Users assessing deployment can compare the models’ license terms, costs and performance on their own workloads as those details become available. The cited global ranking may also change with future model releases and evaluations.
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 much did Super VL improve over its predecessor?
Artificial Analysis scored Super VL at 26, or 26.5 unrounded, compared with 13 for Nemotron 3 Super. It also reported about 49,000 output tokens and $0.25 per index task for Super VL, versus about 114,000 tokens and $1.64 for its predecessor.
Does Super VL lead all open-weight models?
No. Artificial Analysis ranked it 18th among the open-weight models it listed, and 11th among permissively licensed models. The source says models from China and the UAE scored higher.
How does Bolt compare with Nemotron 3 Ultra?
Bolt scored 21, compared with Ultra’s 23, despite having 74 billion total parameters and nine billion active versus Ultra’s 550 billion total and 55 billion active. Bolt had no public API listed in the source, so its API price was not available for comparison.
What is known about Super VL’s license and quantized version?
The source says Super VL uses OpenMDW-1.1, which allows commercial use. Artificial Analysis reported that its NVFP4 version scored within 0.1 points of the BF16 version overall, with no individual evaluation differing by more than 2.6 points.
Source: ThorstenMeyerAI.com
