NVIDIA: Nemotron 3.5 Super VL (free)

Nemotron 3.5 Super VL is an open multimodal mixture-of-experts model from NVIDIA. It belongs to the Nemotron family and is positioned as a vision-language model, with 12B active parameters out of 120B total. On api14 it is listed under the model id nvidia/nemotron-3.5-super-vl:free and its main category is text, with text as the only category tag. The provider release date is 2026-10-15.

The model accepts text, images, and video as input, and it produces text as output. Its context window is 1,000,000 tokens, and the maximum output length is 131,072 tokens. Because it is a mixture-of-experts design with a relatively small active parameter count, it is built to handle large multimodal inputs without activating the full parameter set on every token. The combination of a very large context and image plus video input means a single request can carry long documents, screenshots, or video alongside the prompt.

Those capabilities point to a few practical use cases. Document and screenshot understanding is a natural fit: you can send scanned pages, PDFs rendered as images, or GUI screenshots and ask for extraction, summarisation, or question answering over them. Video input supports tasks such as describing footage, answering questions about events in a clip, or pulling structured data out of recorded sessions. The long context is useful when the material is large, for example a long report plus several images, or a lengthy transcript that needs to stay in one request. Tool calling and structured outputs make it usable as the reasoning step inside an agent or a data extraction pipeline.

You call it through api14's OpenAI-compatible API using the model id "nvidia/nemotron-3.5-super-vl:free". The supported request parameters are include_reasoning, max_tokens, reasoning, response_format, seed, structured_outputs, temperature, tool_choice, tools, and top_p. In practice that means you can control sampling with temperature and top_p, cap the reply with max_tokens, make runs reproducible with seed, request JSON or schema-constrained output through response_format and structured_outputs, and expose functions to the model with tools and tool_choice. The reasoning and include_reasoning parameters let you turn reasoning behaviour on or off and decide whether reasoning content is returned.

One thing to keep in mind is that output is text only, so the model will not generate images or video. If your task is purely text and does not need vision, a text-only model may be simpler, but for mixed text, image, and video input with a very large context, this model covers that ground directly.