Qwen2.5-72B

Qwen2.5-72B is the largest open-weight release of Alibaba's Qwen2.5 series, launched at the Apsara Conference on September 19, 2024. It was trained on 18 trillion tokens with support for up to 128K tokens of context, positioning it as a strong open alternative to frontier proprietary models for long-context and multilingual work.

License nuance: unlike the smaller Qwen2.5 sizes that ship under Apache 2.0, the 72B model is distributed under Alibaba's custom Qwen License Agreement. Commercial use is permitted, but the license is not OSI-approved, so teams with strict open-source requirements should read the terms. It is a capable general-purpose model, though vendor benchmarks should be treated with the usual caution until independently verified.