Intern-Decision 4B runs locally through standard multimodal APIs
Intern-Decision 4B gives developers a locally runnable image-and-text model they can serve through vLLM or SGLang, instead of wiring a custom inference stack. The page shows the model taking an image plus a question, then returning an answer, and provides an OpenAI-compatible chat endpoint for each server. It can also load directly with Transformers using AutoProcessor and AutoModelForMultimodalLM.