NVIDIA Nemotron Can Now Read Your Business Documents For You

NVIDIA released Nemotron 3.5 Super VL on 15 October 2026, an open AI model built to read documents, receipts, charts and video for AI agents instead of only chatting in text. Early users Ramp and Navan are already running it in production, and because its weights are open, any business with the right team can download and host it themselves.

What is NVIDIA Nemotron 3.5?

Nemotron 3.5 Super VL is a 120 billion parameter model that only activates 12 billion of those parameters for any given task, which is why NVIDIA says it runs up to 5 times faster on a single GPU than other open models its size. It reads a 1 million token context window, roughly a few hundred pages at once, and is built specifically for document intelligence: tables, charts, page layouts and scanned text inside PDFs and reports, not just plain sentences. NVIDIA released it under an open commercial license alongside two smaller edge versions, so a developer can choose a bigger model for heavy document work or a lighter one for fast, everyday replies.

Open models can help us move faster, maintain control, and deliver meaningful value without compromising quality.from Ilan Twig, Co-founder and Chief Technology Officer, Navan (Business Wire, 16 October 2026)

What can Nemotron 3.5 actually do?

• Reads a photographed receipt, invoice or form and turns it into structured data such as merchant, date and amount, ready to drop into a spreadsheet.
• Understands tables, charts and page layouts inside a PDF or report, not just the plain text running through it.
• Follows multi-step instructions while calling outside tools, so an agent built on it can look something up, check it and act, rather than answering once and stopping.
• Keeps working across long tasks and video inputs without resetting, which suits an agent left running in the background all day.
• Runs on a single GPU at up to 5 times the throughput NVIDIA reports for comparable open models, which lowers the cost of running it continuously.

How could your business use it?

An accounting firm no longer needs someone to manually key in numbers from a shoebox of receipts, since a model like this can read the photo and output the merchant, date and amount straight into a spreadsheet. A retailer processing supplier invoices could use the same approach to catch a pricing error before a bill gets paid, instead of finding it a month later during reconciliation. A company travel desk booking flights and hotels for staff, the exact job Navan automated, could let an agent check itineraries and receipts against policy without someone combing through PDFs by hand. None of this needs NVIDIA's exact model or an engineering team: the lesson for most business owners is that AI can now read a document about as reliably as it can hold a conversation, which changes which paperwork still needs a human's first look.

Also today

• Meta began charging for WhatsApp Business Platform service messages from 1 October 2026, with roughly 1,000 free messages per business number each month before per message charges apply (AiSensy).
• Malaysia's Budget 2027, tabled 9 October 2026, sets aside RM30 million through MDEC to help 4,000 micro, small and medium businesses adopt AI and automate their operations, alongside training for 5,000 AI professionals (Microsoft News Asia).

What should you try this week?

Pick one stack of paper your business still processes by hand, receipts, delivery notes or claim forms, and try photographing one this week and asking Claude or ChatGPT to pull out the key fields as a table. You will not get NVIDIA's exact speed, but you will see the same shift: AI reading your documents instead of someone typing them out. We teach exactly this kind of practical document and workflow automation, hands on, in Master Claude AI for Business Productivity, where Sabah business owners build their own Claude Skills and Cowork workflows instead of only watching a demo.