Qwen's compact vision-language model gets a 4-bit AWQ build, cutting memory to run multimodal reasoning, OCR, and agent tasks on modest GPUs.
document fetched 2026-10-27 02:55