Ant Group's Bailing announced the open-source release of Ling-3.1-flash, with approximately 560B total parameters, 25B activated parameters, and up to 1M context support. It scored 41 on the intelligence index and outputs about 210 tokens per second. Using MXFP4 quantization and block-FP8, weight usage drops from about 534GB to 292GB, and Hopper deployment is reduced from 8 to 4 cards. It also provides long-context parallel inference and JSON Schema tool-calling support.