ModelRadar/Nemotron 3 Nano Omni (30B)
◆ Underrated

Nemotron 3 Nano Omni (30B)

NVIDIA🇺🇸 US
reasoningmultimodal

NVIDIA's small multimodal reasoning model: 30B MoE, only 3B active, text+image+audio input. Free on OpenRouter. Designed for on-device, 4x faster than its predecessor, reasoning ON/OFF mode.

At a glance

Context
256K tokens
Parameters
30B MoE (3B active)
Input
text, image, audio, code

Specifications

Status
Released
Date
April 30, 2026
Lab
NVIDIA
Origin
US
Type
reasoning, multimodal
Open weights
Yes
License
NVIDIA Open Model License
Context
256K tokens
Parameters
30B MoE (3B active)
Input
text, image, audio, code
Output
text, code

Availability

Direct API access✓ Yes
OpenRouterYes
Open weights✓ Yes