AURA 2.0 Beyond Answering: Learning to Speak First and Act in Streaming Video

Project PageCode on GitHubTechnical Report (arXiv)

AURA 2.0 is a streaming video language model that can initiate conversations, call tools based on the live stream, and ground interactions in objects and regions in video frames.

For demos and source code, see the project page and GitHub repository.

This repository provides the AURA 2.0 checkpoint in compressed-tensors format.

βš–οΈ License and Attribution

The AURA 2.0 weights in this repository are released under the Creative Commons Attribution-NonCommercial 4.0 International license (CC BY-NC 4.0), to the extent of the rights held by the AURA 2.0 team. The weights are intended for non-commercial research. This license does not grant rights to third-party datasets or their underlying media.

Citation

@misc{li2026aura2,
  title={Beyond Answering: Learning to Speak First and Act in Streaming Video},
  author={Li, Shuhan and Guan, Huankang and Guo, Xintong and Bo, Yang and He, Yuyang and Xu, Dunyuan and Chen, Jingyan and Liu, Fang and Wan, Tsz Kin and Chen, Jinpeng and He, Zhenqi and Hu, Yuxuan and Li, Wenbo and Gao, Han and Li, Qijun and Chen, Long and Li, Hongsheng and Liu, Rui},
  year={2026}
}
Downloads last month
37
Safetensors
Model size
9B params
Tensor type
BF16
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for aurateam/AURA_2

Finetuned
Qwen/Qwen3.5-9B
Finetuned
(1,433)
this model

Space using aurateam/AURA_2 1

Paper for aurateam/AURA_2