Ling-3.2-flash

Ling-3.2-flash is Ant Group’s Ant Ling (InclusionAI) latest native multimodal model, launched on 2026-11-05. Changelog: https://developer.ant-ling.com/en/docs/getting-started/changelog ; model docs: https://developer.ant-ling.com/en/docs/models/ling/ .

Positioning: a native multimodal MoE for complex real-world tasks, treating image/text/video as first-class reasoning inputs rather than bolted-on vision modules.

Highlights:

• Scale: 560B total / 25B active, context up to 1M;
• Native image + text + video input;
• Four focus areas: complex visual understanding, multimodal long-horizon agents, 3D understanding & interaction, and medical multimodal;
• Vendor graphics claim wins vs Claude Opus 5 (max) on OmniDocBench 1.5, ZEROBench (Sub), and Vision2Web (vendor-reported; independent verification pending);
• Chat UI for online trials; API compatible with OpenAI and Anthropic protocols; free for the first 2 weeks after launch.

Problem

Many multimodal stacks still treat vision as an external encoder. Ling-3.2-flash folds multimodal perception, deep reasoning, and autonomous execution into one MoE activation budget, with long context and long-horizon agent workloads in mind.

Product & ecosystem

• Latest native multimodal line in Ling 3.2, extending Ling-3.1-flash’s long-horizon agent focus with native vision/video;
• Docs and API shipped as a product drop; third-party paths include OpenRouter examples such as inclusionai/ling-3.2-flash;
• No Hugging Face full open-weight page was found in the same wave as Ling-3.1-flash weights.

Limits & caveats

• OmniDocBench / ZEROBench / Vision2Web comparisons come from vendor release materials;
• Medical multimodal use needs local validation and compliance processes;
• After the free window, platform billing and quotas apply.

Who it's for

• Teams building multimodal agents, document understanding, vision-in-the-loop coding, or medical imaging assist workflows;
• Developers already on Ant Ling’s OpenAI/Anthropic-compatible APIs;
• Groups evaluating a large-context native multimodal MoE.

Getting started

1. Read the changelog and model page: https://developer.ant-ling.com/en/docs/getting-started/changelog and https://developer.ant-ling.com/en/docs/models/ling/ ;
2. Try the Ant Ling chat UI, or call Ling-3.2-flash via OpenAI-/Anthropic-compatible APIs;
3. Use the first-two-weeks free window for load and quality tests;
4. Independently re-run vendor benchmark claims; validate medical or regulated uses offline.