Ling-3.2-flash
Ant Group released Ling-3.2-flash, a native multimodal model with 560B total and 25B active parameters, a 1M-token context window, and text, image and video input.
Ling-3.2-flash adds native vision and video input to the 560B/25B MoE configuration used by Ling-3.1-flash. Ant says it fuses visual and language capabilities at the architecture level instead of attaching a separate vision module. It targets complex visual understanding, multimodal long-horizon agents, 3D understanding and interaction, and medical multimodal work.
- Date
- Thursday 5 November 2026
- Lab
- Ant Group
- Kind
- model
- Access
- closed API
- Price
- $0.30 per million input tokens and $0.90 per million output tokens
Figures
| Measure | Value | Measured by |
|---|---|---|
| Total parameters | 560B 25B activated per step | company |
| Context window | 1M tokens | company |
| OmniDocBench 1.5 | 92.21 vs 90.56 for Claude Opus 5 (max) | company |
| ZEROBench subset | 43.71 vs 39.52 for Claude Opus 5 (max) | company |
| Vision2Web | 85.86 vs 80.20 for Claude Opus 5 (max) | company |
| DesignArena image-to-frontend Elo | 1,238 entered the top 10 | DesignArena |
API is free for the first two weeks, with OpenAI and Anthropic compatible protocols and a chat page. Ant has not said whether weights will be open-sourced; Ling-3.1-flash weights followed a similar free trial under MIT.
Sources
- pandaily.com/ant-group-ling-3-2-flash-native-multimodal-560b-moe-25b-active-1m-context
- technode.com/2026/11/06/ant-launches-ling-3-2-flash-multimodal-model-with-million-token-co
This record was checked against its sources on 6 November 2026. How we check