Ant Ling releases Ling-3.2-flash, a 560B native multimodal model with 25B active parameters
Ant Ling's changelog lists a 1M-token context, native image, text and video input, and a focus on visual understanding, long-horizon multimodal agents, 3D and medical tasks; it is free for two weeks via API, and no weights release was announced.
Why it matters
Ling-3.2-flash extends Ant Group's Ling line from the September Ling-3.0-flash-VL to a natively multimodal mixture-of-experts model, joining the run of large Chinese releases this week alongside Baidu's ERNIE 5.5 Flash Base weights. Its stated focus on medical multimodal tasks and 3D interaction is less common among peers. It is API-only for now with a two-week free period; watch for independent benchmark results and any open-weights release.
Key facts
- 560B total, 25B active parameters
- 1M-token context window
- Native image, text and video input
- Free for two weeks after launch