Ant Group Open-Sources Ling-2.6-Flash Model with Multiple Precision Options

Ant Group’s BaiLing team has open-sourced Ling-2.6-Flash, a 104B-parameter instruction-tuned model.

The model comes in BF16, FP8, and INT4 versions, making deployment more flexible across different hardware setups.

Ant says recent optimizations improved Chinese-English switching, coding compatibility, inference throughput, token efficiency, and agent-oriented performance.

The release is positioned as a lower-cost, faster option for developers building coding workflows and AI agents.