Qwen2.5-72B-Instruct

Qwen2.5-72B-Instruct is the 72B configuration in the Qwen2.5 series. Its dense 72B architecture combines stronger instruction following, longer answers, and structured-data output.

The model card documents 131,072 context tokens and up to 8,192 output tokens. Qwen says the series supports more than 29 languages and was built on up to 18 trillion training tokens. This checkpoint is text-only. Image and video capabilities belong to a VL variant and should not be inferred from the shared Qwen name.