Nemotron 3.5 Super VL is available on Hugging Face
According to the AI Business article, Nemotron 3.5 Super VL is available on Hugging Face along with weights, datasets and training procedures. It processes both documents and video. The sources differ on the availability of the smaller model, Nemotron 3.5 Bolt E4B 4GB.
The AI Business article reports that new models in the Nemotron 3.5 series are available on Hugging Face. Alongside the weights, Nvidia has also published datasets and training procedures. The sources describe the models as open.
The main model, Nemotron 3.5 Super VL, has 120 billion parameters, with 12 billion active per token according to Nvidia. It processes text, images, documents and video and is designed for continuously running enterprise agents. According to Nvidia, it can, for example, link a chart to a table elsewhere in a document and summarize them together without a separate image model. The model is optimized for Nvidia hardware.
The expansion also includes Nemotron 3.5 Bolt 74B-A9B 24GB for local agents. It has 74 billion parameters, with nine billion active, and according to the article in The Decoder, its 2-bit weight quantization allows it to fit into 24 GB of memory. It supports tools such as llama.cpp, Ollama and LM Studio.
The sources differ on the smaller model, Nemotron 3.5 Bolt E4B 4GB: the AI Business article describes it as released and available for conversational agents, while the article in The Decoder reports a later release and a focus on voice assistants.
Why it matters
The published weights and training materials allow developers to customize the models using their own data and run them on their own servers. Optimization for Nvidia hardware is an important consideration when deciding whether to deploy them; for the local variant, 24 GB of memory is a specific prerequisite.
Release card
Nemotron 3.5 Super VL
Nvidia
- Specifications
- The model has 120 billion parameters, with 12 billion active per token.
- Inputs
- The model processes text, images, documents and video.
- Availability
- The model weights are available on Hugging Face. Nvidia has also published datasets and training procedures.
- CharXiv 80.8 % According to Nvidia, the model achieved a score of 80.8 % on a benchmark focused on chart understanding.
- CharXiv s možností spouštět Python 84 % According to Nvidia, the score on the chart understanding benchmark rose to 84 % when the model was allowed to run Python code.
- It is designed for continuously running enterprise agents.
- It can link a chart to a table elsewhere in a document and summarize them together.
- It allows further training on your own company data.
Nvidia claims that the model matches the accuracy of Nemotron 3 Ultra on agent tasks and image analysis with approximately one fifth as many parameters. The company distinguishes it from closed models by the ability to train it further on your own data and run it on your own servers.
The card summarizes information from the article and any dated corrections, with a link to the original source. It is not our assessment of the model. It does not yet have a dedicated editorial profile. Model selection and other announcements →
What was added since the original report
Verified updates
-
According to the article, the models are available on Hugging Face.; The Bolt E4B 4GB variant is designed for conversational agents.; Nvidia says Super VL can complete a task almost instantly.
- According to the article, the models are available on Hugging Face.
- The Bolt E4B 4GB variant is designed for conversational agents.
- Nvidia says Super VL can complete a task almost instantly.
-
12 billion parameters are active per token.; The model processes text, images, documents and video.; Alongside the weights, Nvidia is also publishing datasets and training procedures.; The model is optimized for Nvidia hardware.; The Bolt variant with 4 GB is planned for a later release for voice assistants.
- 12 billion parameters are active per token.
- The model processes text, images, documents and video.
- Alongside the weights, Nvidia is also publishing datasets and training procedures.
- The model is optimized for Nvidia hardware.
- The Bolt variant with 4 GB is planned for a later release for voice assistants.
Two audiences, two different impacts
What this means
For individuals
A developer can try a local agent for writing and debugging code if they have at least 24 GB of memory available for Nemotron 3.5 Bolt 74B-A9B 24GB.
For a business
For automated processing of company documents, Nemotron 3.5 Super VL offers combined analysis of text, tables and charts, along with the option to customize the model using your own data.
ProcessesCheck the original
Event sources
confirmed by 2 independent sources · 3 publishers, 2 independent. We count feeds from the same owner only once.