Ant Group open-sources 560-billion-parameter Ling-3.1-flash AI model

Ant Group has open-sourced the weights of Ling-3.1-flash, a 560-billion-parameter AI model that activates roughly 25 billion parameters per token and supports a context window of up to one million tokens. The release includes deployment support through inference frameworks SGLang and vLLM and targets coding, agent workflows and long-document tasks. Ant Group's mixed-precision deployment approach cuts model-weight memory requirements from about 534GB to 292GB, reducing the reported Hopper GPU requirement from eight GPUs to four.

Ant Group has released the weights of Ling-3.1-flash, making its AI model available for developers to deploy on their own infrastructure. The model has approximately 560 billion total parameters, activates about 25 billion per token and supports a context window of up to one million tokens. The release follows the model’s earlier introduction and includes deployment support through inference frameworks such as SGLang and vLLM. It targets coding, agent workflows and professional tasks that require processing long documents and interacting with external tools. The team has also developed a mixed-precision deployment approach that reduces model-weight memory requirements from approximately 534GB to 292GB. In its reported Hopper GPU configuration, this reduced the deployment requirement from eight GPUs to four. Ant Group https://huggingface.co/inclusionAI/Ling-3.1-flash