Nemotron 3.5 Super VL 120B A12B API Pricing
NVIDIA does not sell Nemotron 3.5 Super VL by the token. Released on October 15, 2026, it is the vision-language member of the Nemotron 3.5 line - text, images, documents and video in, text out - for coding, tool use and visual agents. Its model page has a free evaluation endpoint and full specifications, but no NVIDIA price. The per-token figures on it sit under Available Integrations and belong to hosters: CoreWeave at $0.15/M input and $0.50/M output, Deep Infra at $0.10/M and $0.40/M. Two prices for one model is what a hosting cost looks like; neither is what NVIDIA charges.
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| DeepSeek V4.1 Flash | $0.30 cache $0.006 | $1.20 | $0.15 | 1M | Open weights with vision and a public rate |
| Glimpse | $0.60 | $3.00 | $0.792 | not published | Visual reasoning model with a published rate |
| Solar Mini 4 | $0.10 cache $0.01 | $0.40 | $0.0561 | 512K | Low-cost hosted alternative |
| GLM-5.3 | $1.40 cache $0.26 | $4.40 | $0.78 | 1M | Z.ai's current GLM flagship |
Nemotron 3.5 Super VL 120B A12B has published scores on 2 benchmarks we track; its best placing is #84 of 105 on ARC-AGI-2. All of these scores come from independent leaderboards, not from the lab.
| Benchmark | Score | Rank | Leader | Source |
|---|---|---|---|---|
| ARC-AGI-2 Reasoning | 1.9% ARC Prize verified, best result | #84 of 105 | GPT-6 Astra · 95% | Official leaderboard ARC Prize Foundation |
| ARC-AGI-1 Reasoning | 22.7% ARC Prize verified, best result | #82 of 99 | 6 tied · 98.5% | Official leaderboard ARC Prize Foundation |
Nemotron 3.5 Super VL 120B A12B API pricing FAQ.
What NVIDIA publishes for Nemotron 3.5 Super VL, and what it does not.
Q · 01 How much does Nemotron 3.5 Super VL cost per token? +
Q · 02 But there are prices on the NVIDIA page - why not use those? +
$0.15/M input and $0.50/M output, Deep Infra Serverless at $0.10/M and $0.40/M, with OpenRouter matching Deep Infra. Two prices for the same model on the same page is the clearest sign that the number belongs to the hoster.Q · 03 So how can I use it? +
integrate.api.nvidia.com with the model name nvidia/nemotron-3.5-super-vl-120b-a12b, governed by NVIDIA's API trial terms. A partner endpoint at that partner's rate. Or the open weights, released on Hugging Face the same day under the OpenMDW-1.1 licence, on your own GPUs with vLLM, SGLang or TensorRT-LLM. Unlike earlier Nemotron models, the page lists no downloadable NIM container for it yet.Q · 04 What is the model itself? +
120B total and 12B active parameters, built from Nemotron 3 Super Base. It takes text, images, document pages and sampled video frames and returns text, with configurable reasoning, function calling, structured output and native multi-token prediction for speculative decoding. The model card gives a context of up to 1M tokens shared by input and output; the hosted endpoint's spec panel lists 262K, so check the limit of whichever endpoint you use.Q · 05 What should I price against instead? +
$0.30/M input at peak. That gives you a number to plan with; treat the partner rates above as what hosting Super VL costs today, not as NVIDIA's price.