On September 30, 2026, Ant Group’s Bailing Team officially launched its next-generation large model, Ling-3.1-Flash. Boasting an impressive total of approximately 560 billion parameters, the model activates around 25 billion parameters per token, ensuring robust performance and adaptability. With a maximum context window extending up to 1 million tokens, Ling-3.1-Flash excels in processing extensive documents, complex codebases, and detailed task histories.

Built on a sophisticated Mixture of Experts (MoE) architecture, the model has undergone continuous refinement during its development phase. This optimization process has specifically targeted applications such as general-purpose AI agents, search functionalities, daily office automation, and software development. Additionally, Ling-3.1-Flash has been enhanced to excel in specialized domains, including healthcare, finance, and materials science research.

To encourage widespread adoption, Ling-3.1-Flash will initially offer a two-week free trial period. During this trial, users can access the model with a service length of up to 256K tokens. Following the trial, the model will transition to a paid subscription model, providing users with an expanded 1M-token context window. In a move towards fostering innovation and collaboration, Ant Group also plans to open-source Ling-3.1-Flash simultaneously. Furthermore, the team is committed to ongoing updates aimed at continually improving the model’s performance and capabilities.