inclusionAI/Ling-3.0-flash-VL · Hugging Face

Ling-3.0-flash-VL is a new model that builds upon Ling-3.0-flash, incorporating native image and video understanding capabilities. It features a sparse Mixture-of-Experts (MoE) architecture with 124B total parameters, yet only 5.5B parameters are activated per token, balancing strong multimodal capabilities with inference efficiency. The model supports an extensive context window of up to 1M tokens, enhancing its language, reasoning, and long-context abilities.