Qwen2.5-VL-72B

Qwen2.5-VL-72B is the largest tier of the Qwen2.5-VL vision-language family from Alibaba's Qwen team, released in January 2025. It pairs a 72-billion-parameter language backbone with a redesigned vision encoder supporting dynamic resolution and native video understanding.

The model was positioned as a step change in document parsing, grounding, and long-video comprehension, with the 72B tier serving as the quality ceiling of the family while 3B, 7B, and 32B tiers covered efficiency points.

Qwen2.5-VL is distributed under Apache 2.0, which made the 72B tier one of the largest openly licensed vision-language models at release. It is widely deployed through vLLM, Ollama-class runtimes, and cloud inference providers.