Ant Group ships Ling-3.2-flash, a 560B multimodal model
Ling-3.2-flash pairs 25 billion active parameters with a 1M-token context and native image, text and video input, free for its first two weeks.
Ant Group’s AGI unit, Ant Ling, released Ling-3.2-flash on 5 November: a native multimodal model that takes image, text and video input directly, rather than bolting a vision encoder onto a text model. It’s a sparse mixture-of-experts design with 560 billion total parameters and 25 billion active per token, with a context window of 1 million tokens.
Ant pitches the model at four jobs in particular: complex visual understanding, long-horizon multimodal agent work, 3D understanding and interaction, and medical multimodal tasks. It’s live now through Ant Ling’s own chat interface and through APIs compatible with both the OpenAI and Anthropic protocols, free to use for the first two weeks after launch.
Why it matters: a frontier-scale multimodal model reachable through an API shaped like the OpenAI or Anthropic ones you already integrate against is worth testing during that free window, particularly for agents that need to reason over screenshots, video or 3D scenes rather than plain text. A 1M-token window at this parameter count also makes it a candidate for long-document or long-video agentic work.
The caveat: there are no public weights yet, so this only runs behind Ant’s hosted chat and API, not something you can self-host. The free period is time-limited, and Ant hasn’t published what it charges once that period ends.