Nemotron 3.5 Super VL is an NVIDIA model for tasks requiring several connected reasoning steps, understanding program code or processing substantial material. This Ailoot API listing provides text conversations: you supply instructions, source data and message history, and the model returns a text response.
Use it to analyze technical documentation, compare requirements, prepare code explanations and examine text reports. State the goal, include the necessary material and define the result’s form—for example, a list of conclusions with supporting reasoning, a program-change plan or a comparison of several solutions.
The underlying Super VL is an NVIDIA vision-language model with a C-RADIO vision encoder, also designed to understand documents and video. This Ailoot API listing provides its text capabilities; source material is supplied as text content.
The model’s backbone combines Mamba-2, Transformer and Mixture of Experts architectures. Approximately 12 billion of its roughly 120 billion parameters are active at a processing step. This describes the model’s architecture and should be considered separately from the actual speed of a particular request.
Training used software repositories, task-and-code-change pairs, build logs, multistep problem-solving trajectories and long-context material. Further training combined code, reasoning, instructions and conversation. This specialization is useful when defining a compound task in which source information must first be connected and then organized into a result.
For document analysis, first state the question and information-selection criteria, then include the text with section headings. To compare sources, give them short labels and ask for conclusions to be tied to those labels. For a coding task, add the relevant program fragment, observed behavior and expected result so the response can be checked against a specific case.
A large context window helps place related material in one request. Context size does not replace a clear task definition: separate instructions from source data and specify the decision or document you want to obtain.
For this variant, Ailoot API specifies a window of 1,000,000 tokens and a response limit of up to 131,072 tokens. Conversation history, instructions, source material and the response share the overall context; leave room for the result when preparing a long request.
The model is available through Chat Completions, Responses and Messages. This variant’s available workflow is text input and text output; documents are supplied as text content.
", "aboutText": "### Reasoning, code and long-text work\n\nNemotron 3.5 Super VL is an NVIDIA model for tasks requiring several connected reasoning steps, understanding program code or processing substantial material. This Ailoot API listing provides text conversations: you supply instructions, source data and message history, and the model returns a text response.\n\nUse it to analyze technical documentation, compare requirements, prepare code explanations and examine text reports. State the goal, include the necessary material and define the result’s form—for example, a list of conclusions with supporting reasoning, a program-change plan or a comparison of several solutions.\n\n### Architecture and specialization\n\nThe underlying Super VL is an NVIDIA vision-language model with a C-RADIO vision encoder, also designed to understand documents and video. This Ailoot API listing provides its text capabilities; source material is supplied as text content.\n\nThe model’s backbone combines Mamba-2, Transformer and Mixture of Experts architectures. Approximately 12 billion of its roughly 120 billion parameters are active at a processing step. This describes the model’s architecture and should be considered separately from the actual speed of a particular request.\n\nTraining used software repositories, task-and-code-change pairs, build logs, multistep problem-solving trajectories and long-context material. Further training combined code, reasoning, instructions and conversation. This specialization is useful when defining a compound task in which source information must first be connected and then organized into a result.\n\n### How to prepare a task\n\nFor document analysis, first state the question and information-selection criteria, then include the text with section headings. To compare sources, give them short labels and ask for conclusions to be tied to those labels. For a coding task, add the relevant program fragment, observed behavior and expected result so the response can be checked against a specific case.\n\nA large context window helps place related material in one request. Context size does not replace a clear task definition: separate instructions from source data and specify the decision or document you want to obtain.\n\n### Context and connection methods\n\nFor this variant, Ailoot API specifies a window of 1,000,000 tokens and a response limit of up to 131,072 tokens. Conversation history, instructions, source material and the response share the overall context; leave room for the result when preparing a long request.\n\nThe model is available through Chat Completions, Responses and Messages. This variant’s available workflow is text input and text output; documents are supplied as text content.", "locale": "en", "methods": [ { "html": "POST /v1/messagestexttextPOST /v1/responsestexttextPOST /v1/chat/completionstexttext