Inclusion AI Introduces Ling-3.0-flash-VL with Visual Understanding and Agent Capabilities
Inclusion AI has introduced Ling-3.0-flash-VL, a vision-language extension built upon their Ling-3.0-flash base model. The updated model adds visual perception and multimodal agentic capabilities, performing tasks across STEM reasoning, document intelligence, frontend coding, and medical report interpretation. Adding visual understanding to efficient language models enables developers to deploy visual agent workflows more cost-effectively. It brings capabilities like analyzing UI screens and interpreting complex documents to scalable production environments. The underlying Ling-3.0-flash model uses a Mixture-of-Experts (MoE) architecture with 124 billion total parameters, activating only 5.1 billion parameters per token for fast compute. Ling-3.0-flash-VL builds on this sparse architecture to add multimodal execution capabilities without sacrificing computational efficiency.
## BACKGROUND
Mixture-of-Experts (MoE) is a machine learning technique where only specific subsets of the network parameters (experts) are activated for each token, lowering inference costs while maintaining high capacity. Vision-Language Models (VLMs) combine visual inputs with text processing, allowing AI systems to understand images, diagrams, and user interfaces alongside written instructions.