Ant Ling Ships FP8/FP4/INT4 Quantized Ling-3.0-flash-Fin Checkpoints

Ant Group published FP8, FP4 and INT4 quantized checkpoints of its financial workflow model, Ling-3.0-flash-Fin, on Hugging Face and ModelScope. Instead of shipping a single fixed precision, the release lets a team choose the tradeoff that fits its own constraints, running FP8 where accuracy on financial reasoning matters most or dropping to INT4 where memory and latency are the binding limit on a smaller GPU. That flexibility is particularly relevant for financial deployments, where firms often need to run inference on premises for compliance reasons and cannot simply scale up cloud GPU capacity to compensate for a heavier checkpoint. Quantization at this range is not new on its own, but shipping all three tiers together as first class checkpoints, rather than requiring teams to quantize the model themselves, lowers the engineering cost of adopting Ling-3.0-flash-Fin for a production financial system. It is a practical release aimed squarely at deployment constraints rather than benchmark scores.