native

Definition

Built in from the start rather than bolted on. A natively multimodal model is trained on multiple modalities within one framework from the beginning, as opposed to a text model with a vision encoder attached afterward; native tool use means the capability was trained in rather than prompted around. The word is doing rhetorical work — it claims architectural integration — so it is worth checking which part of the system is actually native.

Explainers using this term

• Paper Explained: SenseNova-U1.5 — One Model That Both Sees and Draws, With No Encoder and No VAESenseNova-U1.5: Towards Native Unified Visual Intelligence
• Paper Walkthrough: Puffin-World — A World Model That Remembers Which Way Is UpPuffin-World: Scaling a Unified Multimodal Model with Native 3D World States
• Paper Walkthrough: DreamX-Creator — Making Sound and Picture Together in 7B, Then Finishing at 2K in One StepDreamX-Creator: Democratizing Native Audio-Video Generation at 2K Resolution