Nemotron 3.5
NVIDIA's open Nemotron 3.5: a 120B Super VL, a 74B Bolt squeezed into 24 GB, and a small E4B. OpenMDW license.
In short. Nemotron 3.5 is NVIDIA's family of open AI models for building agents. Super VL is the big one for servers and reads text, images and video. Bolt is the local one: a 74-billion-parameter version compressed to fit in 24 GB of memory, and a small version for voice assistants. The license is permissive, so commercial use is allowed.
Summary
NVIDIA released Nemotron 3.5 Super VL and Nemotron 3.5 Bolt 74B-A9B on 15 October 2026, and Bolt E4B on 23 October. All three use the OpenMDW-1.1 license, and the model cards say each is "ready for commercial or non-commercial use".
Super VL is a vision-language mixture of experts with 120 billion parameters and 12 billion active per token. It interleaves Mamba-2 layers, attention layers and expert layers, adds a C-RADIO vision encoder, reads text, images and video, writes text, and supports up to 1 million tokens of context. The BF16 weights are about 249 GB, which the card serves on four H100 GPUs, and an NVFP4 version is about 75 GB. NVIDIA says it gives up to five times the throughput of other open models of its size; that is a self-reported claim.
Bolt 74B-A9B 24GB is the local model. It has 74 billion parameters with 9 billion active, was trained with 2-bit quantization so the packed model fits in 24 GB, accepts text, images and video, and supports up to 256K tokens. NVIDIA positions it for everyday agent work on RTX PCs and DGX Spark, with support for llama.cpp, Ollama, LM Studio, Unsloth and vLLM. A full BF16 version of about 150 GB is also published.
Bolt E4B is a small text model that NVIDIA post-trained from Google's Gemma 4 E4B, for voice and device assistants. It has 8 billion parameters in total (about 4 billion effective), a 64K context, and a roughly 4 GB quantized build for constrained devices.
Good for
- Document, screenshot and video understanding inside agent workflows (Super VL)
- Running a capable agent locally on a 24 GB GPU or a DGX Spark-class machine (Bolt 74B-A9B 24GB)
- Voice assistants and embedded interfaces with tight memory budgets (Bolt E4B)
- Teams that want a permissive license and open recipes for fine-tuning
Watch out for
- Super VL is a server model, at 249 GB in BF16 and about 75 GB even in NVFP4
- The 24 GB Bolt is trained at 2 bits to save memory, so expect lower quality than the full-precision 74B version
- The shipped recipes and the NVFP4 files target NVIDIA GPUs; speed on other hardware will differ
- Speed and accuracy claims come from NVIDIA's own tests
Performance claims come from the developer and have not been independently verified.
How much memory does it need?
Estimates for the Bolt E4B variant. A mixture-of-experts model must hold all of its experts in memory even though only some are used for each token, so total parameters decide the memory and active parameters decide the speed.
| Variants | Parameters | BF16 | FP8 / INT8 | 4-bit (Q4) | Fits on |
|---|---|---|---|---|---|
| Bolt E4B BF16 | 8B 4B active | 18 GB | 9.2 GB | 5.52 GB | Laptop or phone |
| Bolt 74B-A9B BF16 | 74B 9B active | 170 GB | 85 GB | 51 GB | Workstation |
| Bolt 74B-A9B 24GB | 74B 9B active | 170 GB | 85 GB | 24 GB | Gaming GPU |
| Super VL BF16 | 120B 12B active | 276 GB | 138 GB | 83 GB | Workstation |
| Super VL NVFP4 NVFP4 | 120B 12B active | 276 GB | 138 GB | 83 GB | Workstation |
Estimates: weights plus about 15% overhead. Real usage grows with context length and depends on the software. The developer states a minimum of 24 GB of GPU memory.
The license in plain words
A short permissive license written for machine-learning models by the Linux Foundation and PyTorch Foundation. It covers the model, its data, documentation and software together, and puts no restrictions on what you do with the model's outputs.
See also
- Best open-source LLMs in 2026
- Giant open models now ship in 4-bit: what it means for memory planning
Frequently asked questions
- What is Nemotron 3.5?
- Nemotron 3.5 is NVIDIA's family of open AI models for building agents. Super VL is the big one for servers and reads text, images and video. Bolt is the local one: a 74-billion-parameter version compressed to fit in 24 GB of memory, and a small version for voice assistants. The license is permissive, so commercial use is allowed.
- Can I use Nemotron 3.5 commercially?
- Yes. Nemotron 3.5 is released under OpenMDW-1.1, a permissive license. You can use it commercially if you follow the standard conditions, such as keeping the license notice.
- Is Nemotron 3.5 open source?
- Nemotron 3.5 is open-weight, not open source in the strict sense. You can download the model files, but the OpenMDW-1.1 adds conditions that the Open Source Initiative definition does not allow.
- Can I run Nemotron 3.5 on my own computer?
- Yes. The smallest variant, Bolt E4B, needs about 5.52 GB of memory at 4-bit precision, which fits a modern laptop or a mid-range graphics card.
- When was Nemotron 3.5 released?
- Nemotron 3.5 was released by NVIDIA on October 15, 2026.
- How long a context window does Nemotron 3.5 have?
- Nemotron 3.5 supports a context window of up to 1M tokens.
Official links
Sources
Specifications come from these primary sources. Benchmark claims, where mentioned, are reported by the developer.
- NVIDIA blog: Nemotron 3.5 Super VL and Bolt (15 Oct 2026) blogs.nvidia.com
- Nemotron 3.5 Super VL BF16 model card (Hugging Face) huggingface.co
- Nemotron 3.5 Bolt 74B-A9B 24GB model card (Hugging Face) huggingface.co
- NVIDIA Research: Nemotron 3.5 Bolt E4B 4GB (23 Oct 2026) research.nvidia.com
Release date and facts last verified November 8, 2026.
Related models
| Model | Released | Parameters | Context | License | Smallest variant needs |
|---|---|---|---|---|---|
| Gemma 4 Google DeepMind | 5.1B – 30.7B | 256K | Apache 2.0 (Permissive) | 3.52 GB Laptop or phone | |
| Nemotron 3.5 Lightning NVIDIA | 31.6B 3B active | 1M | OpenMDW-1.1 (Permissive) | 22 GB Gaming GPU | |
| Nemotron 3 Ultra NVIDIA | 550B 55B active | 1M | OpenMDW-1.1 (Permissive) | 379 GB Big-memory desktop | |
| Nemotron 3.5 Transcribe NVIDIA | 34B 4.3B active | 64K | OpenMDW-1.1 (Permissive) | 25 GB Gaming GPU |