NVIDIA: Nemotron 3 Nano Omni 30B A3B Reasoning (free)

Nemotron 3 Nano Omni 30B A3B Reasoning is NVIDIA’s 30B-total, 3B-active model described for perception and context sub-agent work in enterprise agent systems. The Ailoot method for this edition is a text conversation through POST /v1/chat/completions. Select the exact model ID nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free and provide a nonempty messages array containing the instructions and source text for the task. Each message has a role of system, developer, user or assistant and plain string content. An optional name is a string; keep each message to these fields. Send stream:false, or omit stream.

Use it to examine a supplied body of material and return a focused answer, a comparison, or a context brief for the next step of an application. A handoff prompt can provide the original question, relevant excerpts and constraints, then ask for established facts, unresolved questions and the source passages that support each conclusion. For a comparison, name the criteria and include the alternatives’ actual specifications so the model can work from the same evidence.

This edition accepts text input and produces text output. To work with a document, insert the relevant UTF-8 text into message content and retain headings or source labels that help distinguish the excerpts. Supply the conversation history needed by a follow-up. Keep any tool execution or retrieval in the surrounding application, and pass its completed textual results into the conversation. Media inputs, tool-calling controls and explicit cache controls are outside this edition’s request profile.

The model page lists a 256,000-token context window and up to 65,536 completion tokens. Ailoot also requires the entire UTF-8 JSON request body, including instructions and field names, to fit within 256,000 bytes. For an explicit output cap, set exactly one of max_tokens or max_completion_tokens to a positive integer from 1 through 65,536. If neither is supplied, Ailoot forwards a 4,096-token output cap. Keep the complete supplied conversation and requested answer within the corresponding limits. A long input is useful when the answer depends on many connected excerpts, but a focused selection and a concrete deliverable make it easier to inspect whether the response follows the evidence. Prompted structure can request named sections or a table; treat the returned text as the deliverable rather than assuming a schema-enforced object.

For a complete response, send stream:false and retain the full JSON. Read the generated text from choices[].message.content and keep every returned choice, finish_reason, usage and any supplied reasoning fields. Preserve source labels in the final answer when your workflow needs to trace a conclusion back to its input. Check that the requested answer is complete before passing it to another application step.

Prepare public or synthetic source material for this free edition; its documented usage terms exclude confidential information and personal data. If a response is incomplete or the send outcome is uncertain, keep the original request identifier and response for checking the existing attempt. The text request returns its result through the conversation response; no separate asynchronous job or cancellation route is documented for this card.