Ant Financial's open-source, high-efficiency instruct model, Ling-2.6-flash.

Ant Group's Bailing large-scale model announced the official open-sourcing of Ling-2.6-flash, simultaneously providing multiple precision versions including BF16, FP8, and INT4. The model is already available on Hugging Face and ModelScope. The model has a total of 104 bytes of parameters and 7.4 bytes of activation parameters. Two weeks ago, it anonymously appeared on OpenRouter as Elephant Alpha, quickly gaining widespread attention from the developer community and industry users.