Ant Group announces open-sourcing of multimodal large model Ling-3.0-flash-VL

Ant Group’s open-source announcement has officially unveiled and open-sourced Ling-3.0-flash-VL, the first native multimodal large model in its Ling series. Derived from the MoE (Mixture of Experts) architecture of Ling-3.0-flash, the model has a total parameter count of 124 billion, with 5.5 billion parameters activated per inference. It natively supports image, text, and video inputs, and features a 256K-token context window. Focused on the theme of “how to complete real-world tasks more reliably and efficiently”, Ling-3.0-flash-VL explores key directions including: Integrating visual capabilities into large models— a common concern is that this will diminish text intelligence, but training practices have yielded the opposite result: native multimodal joint training not only expands application boundaries but also enhances text intelligence. The model introduces a visual feedback mechanism: observing execution results, comparing against targets, identifying discrepancies, and iterating corrections, shifting tasks from one-off “generation” to a closed loop of “observe → act → verify → correct” to boost execution reliability. It also inherits Ling-3.0-flash’s core strength as an efficient execution node in Agent workflows, balancing output quality and operational efficiency within the visual feedback loop, enabling full task completion at lower cost and in shorter time.