Inclusion AI releases Ling-3.0-flash-VL, a 124B multimodal MoE with 1M context

Inclusion AI released Ling-3.0-flash-VL, a multimodal extension of its Ling-3.0-flash language model that adds native image and video understanding while keeping the base model's reasoning and long-context abilities. The model carries 124B total parameters but activates only 5.5B per token through a sparse mixture-of-experts architecture, and supports a context window up to 1M tokens.

The design routes visual input through a ViT encoder whose features are aligned to text representations by a two-layer MLP projector. VideoRoPE encodes spatial position and temporal order so the model can track visual change over time, which the company says supports event localization, long-video question answering, and video clip editing. A 42-layer hybrid backbone alternates KDA and Gated MLA layers at a 5:1 ratio for efficient long-context processing across text, images, video, and extended agent task histories.

Inclusion AI frames the release around real-world reasoning and agentic workflows rather than chat alone. The model is listed on Hugging Face, though no benchmark results or deployment details were included in the announcement.