Text model

ERNIE 4.5 VL 28B A3B Thinking

A multimodal MoE model that “looks, reads, and reasons” across images, video, and text. It adds tool use and a Thinking with Images mode, supports long context, and activates about 3B parameters per token for flagship-level VLM quality at practical latency.

Baidu Text Closed apache-2.0
Type
Text
Weights
Closed
Downloads
684
Likes
543
Pipeline
image text to text
License
apache-2.0