Qwen2 Instruct 72B

Qwen2 Instruct 72B is an instruction-tuned large language model released by Alibaba as part of the Qwen2 family. Within that generation, the 72B dense model sits at the top of the parameter range, above the smaller Qwen2 sizes that Alibaba shipped alongside it. The "Instruct" designation indicates the chat and instruction-following variant rather than the base pretrained checkpoint, so it is the version typically exposed through hosted inference APIs and used directly for assistant-style workloads. As a Qwen2-generation model, it is aimed at general-purpose text generation: multi-turn dialogue, reasoning over instructions, code generation and explanation, and multilingual work, with Chinese and English as the most prominent languages in the Qwen series. Because it is a dense 72B model, serving it requires substantially more accelerator memory than the smaller Qwen2 checkpoints, which is why hosted API access is a common route for teams that do not want to provision multi-GPU nodes themselves. We do not track a confirmed context window, modality list, or benchmark scores for this entry, so treat provider documentation as the authority on the exact context length and feature set each host exposes. In practice, Qwen2 Instruct 72B is used where teams want a large open-weight-style alternative to proprietary mid-to-large models, particularly for Chinese-language or bilingual applications. It has since been followed by later Qwen generations, so it is generally chosen today either for continuity with existing deployments, for cost reasons relative to newer large models, or where a specific provider offers it. Compare the providers listed in the pricing table on this page, since throughput and context limits vary between hosts of the same weights.