On November 7, Ant Group's Ling platform went live with Ling-3.2-flash, a native multimodal model with 560 billion total parameters that takes image, text and video as input.
Key points
- 560 billion total parameters, 25 billion active per token
- 1 million token context window
- Three input modalities: image, text and video
- Free for two weeks during the launch trial period
It supports a visual self-feedback error-correction loop that retries when a tool call fails.
Read the full story on threerouter: https://www.threerouter.com/blog/358
