NVIDIA Nemotron 3.5 Super VL API: images, PDFs and tool calls

NVIDIA's Nemotron 3.5 Super VL is available on FastMetal as nemotron-3.5-super-vl, through the API and in chat. It is an open-weight vision-language model that handles reasoning, coding, tool use and document and image understanding in one model.

Capabilities

• Image and PDF input. Send images alongside text, or pass a PDF as is. We tested both with real files before launch.
• Tool calling and structured output. Function calling (tools) and responses constrained to a JSON schema both worked on real requests.
• 120B parameters, 12B active. It is a hybrid Mamba-2 and Transformer mixture of experts, so only 12B parameters run per token. NVIDIA designed it to run light for its size.
• A 262K-token context. We serve it with 262,144 tokens of context. The model itself supports up to 1M.
• Adjustable reasoning. It thinks before answering, and you can turn thinking on or off and set how deep it goes.
• Fewer confident wrong answers. NVIDIA reports a hallucination rate of 30.0% on AA-Omniscience, against 84.5% for Qwen3.5-122B-A10B and 95.9% for DeepSeek-V4-Flash.
• Open weights under the OpenMDW-1.1 license. We serve NVIDIA's own NVFP4 release on a host that does not train on prompts.

Pricing

Nemotron 3.5 Super VL is billed per million input tokens and per million output tokens, with cached input at a lower rate. Thinking tokens count as output. Current rates are on the model page.

Billing is prepaid: you load a balance by card, from $5, and you can never spend more than you loaded. See pricing.

Calling it

Use the OpenAI-compatible Chat Completions API with nemotron-3.5-super-vl as the model. This request asks it to describe an image:

curl https://api.fastmetal.ai/v1/chat/completions \
  -H "Authorization: Bearer $FASTMETAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "nemotron-3.5-super-vl",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Describe what is in this image."},
        {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
      ]
    }]
  }'

To try it without code, pick Nemotron 3.5 Super VL in chat and attach an image or a PDF to your question.

model: "auto" does not pick it for now; call it by name.

Compare real answers first

The model page shows its actual answers to our 43 shared prompts, from summaries to bug fixes to translation. Compare them with the previous generation, Nemotron 3 Super, or any other model in the catalog.