Ant Group Announces Ling-3.1-flash Model with 560B Parameters and Open-Source Plans
Ant Group's Bailing team has released Ling-3.1-flash, a Mixture-of-Experts model boasting approximately 560 billion total parameters, 25 billion active parameters per token, and a context window of up to 1 million tokens. The model is available for a two-week trial with a temporary 256K context limit, after which Ant Group plans to enable the full 1M context and open-source the model. Ling-3.1-flash highlights how modern MoE architectures enable high model capacity while keeping per-token computational costs low during execution. Ant Group's commitment to open-sourcing the model post-trial will provide the global open-source community with access to a competitive ultra-long context language model. By activating only about 25 billion parameters per token out of 560 billion total, the model optimizes compute efficiency for long-context tasks across financial, medical, and software engineering domains. The trial limits context length to 256K tokens to manage operational costs before expanding to 1M tokens upon full commercial release.
## BACKGROUND
Mixture-of-Experts (MoE) is a model architecture that routes input tokens to selective sub-networks rather than processing them through the entire network, decoupling parameter size from computational cost per request. Ant Group's Ling series encompasses its independently developed foundation models, which are tailored for domain-specific AI agents, search, and enterprise applications.