How much memory does the local Bolt model need?

Nemotron 3.5 Bolt 74B-A9B 24GB needs 24 GB of memory, which Nvidia achieves with aggressive 2-bit quantization of the weights.

What does Nvidia release alongside the model?

Nvidia is releasing the weights in base, post-trained, and 4-bit compressed versions, along with datasets, training recipes, and reinforcement learning environments.

How is Super VL trained for Indian languages?

Nvidia built a dataset with the Indian Institute of Technology Madras and its AI lab AI4Bharat containing about 2.83 trillion tokens of text in 22 Indian languages.

Large Language ModelsOpen-Source AIAI Business & IndustryTHE DECODER4m agoOct 16, 2026, 04:00 JST

Nvidia ships Nemotron 3.5 Super VL, a 120 billion parameter open model

Send on LINE

3 Key Points

What happened
Nvidia released Nemotron 3.5 Super VL, an open multimodal model with 120 billion parameters for enterprise agents, plus two compact Bolt models for local PCs and edge devices.
Why it matters
Nvidia says the model matches its largest reasoning model's accuracy at one-fifth the parameters and can be fine-tuned on customer data and run on customer servers, which may appeal to insurers and financial firms.
What to watch
The smaller Bolt E4B 4GB, aimed at fast voice assistants, is set to follow later this month.

WHO IT HITSEnterprise IT teams that want to fine-tune models on sensitive data and run them on their own servers, plus developers and owners of high-end workstations with 24 GB of memory who want to run agents locally.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

Nvidia is positioning Nemotron 3.5 Super VL as a model that enterprises can fine-tune with their own data and run on their own servers, free from a provider's terms of service. That pitch targets industries with heavy document loads and sensitive data, such as insurance claims screening and credit scoring.

Early adopters span finance and cybersecurity. Scale AI adapted the model for vendor credits at Skydance, Ramp uses it to turn receipt photos into structured data, and AI21 fine-tuned it for security firm Alice, which Nvidia says improved accuracy by 40 percent.

The strategy behind Nvidia's open models is self-serving. The more companies run their own open models, the greater the demand for compute hardware. Nvidia recently invested $12.9 billion in Hugging Face, one of the most important platforms for developing and sharing open models.

FAQ

How much memory does the local Bolt model need?: Nemotron 3.5 Bolt 74B-A9B 24GB needs 24 GB of memory, which Nvidia achieves with aggressive 2-bit quantization of the weights.
What does Nvidia release alongside the model?: Nvidia is releasing the weights in base, post-trained, and 4-bit compressed versions, along with datasets, training recipes, and reinforcement learning environments.
How is Super VL trained for Indian languages?: Nvidia built a dataset with the Indian Institute of Technology Madras and its AI lab AI4Bharat containing about 2.83 trillion tokens of text in 22 Indian languages.

THE DECODERRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.