NVIDIA: Nemotron 3 Nano Omni (free)

NVIDIA Nemotron 3 Nano Omni is an open multimodal model from NVIDIA, released on 28 April 2026. It is a 30B-A3B model, meaning it uses a mixture-of-experts design with roughly 30 billion total parameters and about 3 billion active per token. NVIDIA positions it as a perception and context sub-agent for enterprise agent systems, so it is intended to slot into larger pipelines rather than act as a standalone assistant.

On input, the model accepts text, images, audio and video, which makes it unusual among text-focused models. Output is text only. The context window is 256,000 tokens, and the maximum output is 65,536 tokens. It supports reasoning behaviour, including a reasoning parameter and an include_reasoning option, so you can request that the model's step-by-step reasoning be surfaced or kept separate depending on your application.

Those capabilities point to a fairly specific set of use cases. Because it handles video and audio as well as images, it fits tasks like summarising recorded meetings, extracting structured data from screenshots or scanned documents, captioning or indexing media libraries, and answering questions over mixed-media content. Its long context helps when a single request must cover a long transcript, a batch of images, or a combination of the two. As a sub-agent, it can feed perception results into a larger orchestration layer, for example converting raw media into text that another model then reasons over.

You call it through api14's OpenAI-compatible API using the model id "nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free". Because the endpoint follows the OpenAI chat completions shape, existing SDKs and clients work with minimal changes. The supported request parameters are include_reasoning, max_tokens, reasoning, seed, temperature, tool_choice, tools and top_p. The presence of tools and tool_choice means you can define function calls and let the model decide when to invoke them, while seed, temperature and top_p give you control over sampling consistency.

One practical point to weigh is scope. The model outputs text only, so if you need generated images, audio or video, it is not the right fit. It is also a Nano-class model, and NVIDIA describes it as a sub-agent, so for tasks demanding the strongest possible reasoning you may prefer a larger frontier model and use Nemotron 3 Nano Omni for the perception and context stage. The free tier carries the usual caveat that a free variant may see different capacity or availability than paid endpoints, so test it against your expected load before depending on it in production.