🔍 Read the full analysis: Nemotron 3.5 Doubles NVIDIA’s Score. America’s Open-Weight Gap Is Still There. on ThorstenMeyerAI.com
Get business pricing on tech for your team
- 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. Artificial Analysis scored Super VL at 26, twice the score of its predecessor, but the highest-scoring open-weight models still come mainly from China, with one leading model from the UAE.
NVIDIA released two open-weight reasoning models on October 15, and an independent evaluation by Artificial Analysis scored its Nemotron 3.5 Super VL at 26, double the score of its predecessor seven months earlier. The release strengthens the United States’ position among commercially permissive models, but the highest scores on the broader open-weight leaderboard still belong to models from China and the United Arab Emirates.
Artificial Analysis reported a score of 26 for Super VL, or 26.5 before rounding, on its Intelligence Index. The previous Nemotron 3 Super scored 13. The two models are in the same size class, with 120 billion total parameters and 12 billion active parameters. The newer model used about 49,000 output tokens per index task, compared with roughly 114,000 for its predecessor. Artificial Analysis put the cost per task at $0.25, down from $1.64.
NVIDIA also released Nemotron 3.5 Bolt 74B-A9B, which scored 21. It has 74 billion total parameters and 9 billion active parameters. Artificial Analysis ranked it fourth among models under 100 billion total parameters and first among US models in that size group. Bolt has no public API yet, so the source evaluation did not report an API price or task cost.
Super VL accepts text and images and returns text, and its OpenMDW-1.1 license allows commercial use. NVIDIA provides it in NVFP4, a four-bit format. Artificial Analysis found that version scored within 0.1 points of full-precision BF16 weights in its evaluation, with no individual evaluation differing by more than 2.6 points. The source reports an API price of $0.20 per million input tokens and $0.575 per million output tokens.
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 US Option, Still Behind
The results give developers a higher-scoring US model under a license that permits commercial use. Super VL’s combination of score, reported task cost and reduced token use may matter to organizations comparing deployed models, while its NVFP4 version could lower memory needs for self-hosting. Those figures come from Artificial Analysis’s evaluation; deployment speed and cost can vary by provider and hardware.
The broader ranking limits what the release changes. Artificial Analysis lists Super VL at 18th among open-weight models, or 11th when counting only permissively licensed models. The source says its highest-scoring permissive open-weight model, Xiaomi’s MiMo-V2.6-Pro, scored 46. Super VL is a measurable gain for NVIDIA, but it does not put a US model at the top of the overall ranking.
Products Worth Considering
Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB‑C Charging, Up to 2X More Active Noise Cancelling, Transparency Mode, Adaptive
Apple Watch SE 3 GPS 40mm Smartwatch
Bose Smart Soundbar 700 with Alexa, Black
As an affiliate, we earn on qualifying purchases.
How the Leaderboard Is Counted
Artificial Analysis evaluates models with its Intelligence Index and reports estimated costs for completing an index task. Those results provide a common comparison, but they do not by themselves establish performance across every workload. Provider, prompt length and serving setup can affect speed and actual cost.
License terms also change which models count as commercially permissive. The source describes Meta’s Muse Spark Open, released on the same day, as scoring 39 and using a license with commercial restrictions. It says nine of the ten highest-scoring open-weight models listed by Artificial Analysis came from Chinese labs, with the tenth being Muse Spark Open. The UAE’s K2 Horizon 375B A23B scored 31 and uses Apache 2.0, according to the source.
Limits of the Current Comparison
The supplied material does not include the full evaluation methodology or enough detail to independently assess how the index maps to particular customer workloads. The reported speed of about 630 output tokens per second came from a pre-release DeepInfra endpoint serving NVFP4 weights; the source says the median across six providers listed on the model page was 196 tokens per second. Results may differ across providers.
Bolt has no public API, and Artificial Analysis had not listed a license for it in the supplied material. Its cost per task is therefore not available. The source also begins a comparison of Muse Spark Open’s license but does not provide its complete terms, so the specific commercial restrictions cannot be detailed here.
API Access and Rankings Ahead
The next practical comparison for Bolt depends on public access and pricing: without an API or a reported task cost, buyers cannot compare its end-to-end economics with Super VL. For Super VL, provider-level serving results will show how closely users can reproduce the reported speed and cost figures in their own deployments.
Artificial Analysis’s leaderboard may change as models and evaluations are added. The current results establish that NVIDIA improved its own score and gained ground among permissively licensed US models; they do not establish that the US has closed its overall open-weight gap.
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 reasoning models that accept text and images and return text.
How much did Super VL score?
Artificial Analysis scored Super VL at 26 on its Intelligence Index, or 26.5 before rounding. Its predecessor scored 13 in the comparison supplied.
Does Super VL lead all open-weight models?
No. The source places it 18th among all open-weight models listed by Artificial Analysis and 11th among permissively licensed models. It describes Xiaomi’s MiMo-V2.6-Pro as scoring 46.
What is known about Bolt’s cost?
Artificial Analysis did not report a cost per index task for Bolt, and the model had no public API in the supplied material. Its deployment cost remains unclear.
Source: ThorstenMeyerAI.com
