Nemotron 3 Nano Omni 30B A3B Reasoning

Nemotron 3 Nano Omni 30B A3B Reasoning is positioned within the broader Nemotron family as NVIDIA's first multimodal entry in the Nano line, extending the series beyond text-only reasoning into a unified perception-and-language framework. The DeepInfra launch announcement confirms the model's public release as a coordinated effort between NVIDIA and its hosting partners, signaling an intent to bring capable multimodal reasoning into accessible inference infrastructure rather than restricting it to research deployments.

Because the supplied excerpts contain only navigation chrome and high-level release confirmation, the model's specific architectural strengths, benchmark standing, and real-world fit must be inferred from its stated purpose: serving as a compact, reasoning-oriented multimodal model that can ingest diverse inputs and produce coherent text outputs. For practitioners, this suggests a model suited to integrated workflows where understanding across text, image, video, and audio matters more than raw scale, particularly in agentic pipelines that benefit from combined perception and deliberation in a single model.

Requestynemotron-3-nano-omni-30b-a3b-reasoningnemotron

Open provider page Provider docs

Quick Info

Provider: Requesty
Model key: nemotron-3-nano-omni-30b-a3b-reasoning
Release date: Apr 28, 2026
Last updated: Apr 28, 2026
Input modalities
Output modalities
Capabilities

Cost

A provider subscription or plan supersedes token-based pricing for this model.

Limits

Output tokens: 20,480 tokens
Context window: 131,072 tokens

Latest news about Nemotron 3 Nano Omni 30B A3B Reasoning

Requesty

CoverageBenchmark

Artificial Analysis provides an independent technical assessment of NVIDIA's Nemotron 3 Nano Omni 30B A3B Reasoning, an open-weights model released in April 2026. The model scores 10 on the Artificial Analysis Intelligence Index, placing it above average among comparable models (class median: 8). Pricing is reported at Specifications confirm the model is a reasoning-capable variant with 30B total parameters and 3B active parameters per token, built on a hybrid MoE Transformer-Mamba architecture per cross-corroborated metadata. It supports multimodal input (text, image, speech, video) with text output and a 256K token context window (

Requesty

Official sourceBenchmark

Requesty lists NVIDIA's nemotron-3-nano-omni-30b-a3b-reasoning as a managed deployment served directly from NVIDIA in the US, exposed via the OpenAI-compatible endpoint at https://router.requesty.ai/v1 using the model id nvidia/nemotron-3-nano-omni-30b-a3b-reasoning. According to the page, the endpoint was added in Jun The model is described as a multimodal MoE unifying video, audio, image, and text understanding, with chain-of-thought reasoning and tool calling aimed at enterprise Q&A, summarization, transcription, OCR, GUI automation, and document intelligence. Endpoint reference specs give a 131K-token context window and 20K-token