33.7B
- Context
- 1Mtokens
- Released
- 4w ago
- Size
- 120Bparameters
Nemotron 3.5 Super 120B A12B is an open vision-language model from NVIDIA and part of the Nemotron family, initialized from Nemotron 3 Super and post-trained into a multimodal agent. NVIDIA's launch announcement presents it as the fastest and most accurate open model for always-on agents, with a claimed up-to-5x throughput gain on agentic workloads compared with similar-size open models. It is built for agentic work such as reasoning, coding, tool use, and document and video understanding.
- Language model
- Open weights · 120B
- 1M context
- Reasoning
- Tools
How it compares
Size class
Large120BAn open-weight model of 100 billion parameters or more, which needs a multi-GPU server.
Open models of a similar size from other labs
Vs. your stack
Add the models you use to see how this one compares.
Sign inAdditional source evidence
- Source model
- nvidia/nemotron-3.5-super-120b-a12b
- Source family
- nemotron
- Last updated
- 2026-10-15
- Model context
- 1M
- Model output limit
- 131K
- Open weights
- Yes
- File attachments
- No
- Reasoning
- Yes
- Tool calling
- Yes
- Temperature control
- Yes
Complete source record (JSON)
{
"id": "nvidia/nemotron-3.5-super-120b-a12b",
"name": "Nemotron 3.5 Super 120B A12B",
"description": "NVIDIA Nemotron 3.5 Super reasoning model (120B total, 12B active) served with text-only input",
"family": "nemotron",
"last_updated": "2026-10-15",
"attachment": false,
"reasoning": true,
"tool_call": true,
"temperature": true,
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"limit": {
"context": 1048576,
"output": 131072
},
"release_date": "2026-10-15",
"open_weights": true
}Frequently asked questions
Running locally
An estimated 84.2 GB of VRAM at Q4, 147.1 GB at Q8 and 276.5 GB at FP16, derived from the 120B parameter count. All figures are estimates: weights plus about 15% and 0.5 GB for a short context.
No. By estimate, Nemotron 3.5 Super 120B A12B does not fit on a 24 GB, 16 GB, 12 GB or 8 GB GPU at Q4, Q8 or FP16.
Yes, but only on large-memory Macs. The smallest common Mac memory size that fits the estimated 84.2 GB needed at Q4 is 128 GB, assuming macOS lets the GPU use about 75% of unified memory.
Privacy and data
Yes. Nemotron 3.5 Super 120B A12B has open weights, so you can self-host it for full control of your data.
Capabilities and specs
The context window is 1M tokens (1,048,576) and the maximum output is 131K tokens (131,072).
No. The only input modality is text, so it cannot read images, audio, video or PDFs.
Yes, tool calling (function calling) is supported.
Yes, a reasoning mode is supported.
Quality and benchmarks
Kolibri 1 by Aleph Alpha, ERNIE-5.5-Flash-Preview by Baidu and Flint by Recursive. They were matched on model kind and release date, with no comparable price.
Kolibri 1Nemotron 3.5 Super 120B A12B vs Kolibri 1ERNIE-5.5-Flash-PreviewNemotron 3.5 Super 120B A12B vs ERNIE-5.5-Flash-PreviewFlintNemotron 3.5 Super 120B A12B vs Flint
Licensing and access
Open weights. Nemotron 3.5 Super 120B A12B is released as open weights.
Basics
NVIDIA made it, releasing it on October 15, 2026.
Open weights. The facts state only that Nemotron 3.5 Super 120B A12B has open weights, with no pricing information given.
Generated from Vertical Tabs data, updated November 12, 2026.
Discussion
No comments yet
Be the first to share your thoughts.