Ant Group's Bailing Big Model Ling-2.6-1T Officially Open-Sourced, Focusing on Complex Task Optimization

Ant Group's Bailing big model announced today the official open-source release of Ling-2.6-1T. Instead of pursuing parameter scale or long thought chains, the model focuses on real complex tasks, systematically optimizing efficiency, instruction execution, tool adaptation, and other capabilities.

Ling-2.6-1T uses hybrid architecture innovation and reinforced reward strategies to maintain 1T parameter capabilities while reducing Token overhead. In multi-step execution scenarios like Agent and Coding, it achieves open-source SOTA levels on multiple execution benchmarks. It also has complete engineering implementation capabilities, is compatible with mainstream Agent frameworks, and can integrate into developers' and enterprises' production workflows.

The model completed the full Artificial Analysis evaluation with 16M tokens. The official also extended the free API call service on the OpenRouter platform for one week to facilitate more developers' experience.