Qwen2

Qwen2 is the June 2024 generation of Alibaba's Qwen large language model family, released in five sizes from 0.5B to 72B parameters, including a 57B mixture-of-experts variant. Pretrained on 7 trillion tokens with support for 29 languages beyond English and Chinese, the 7B and 72B instruction models handle up to 128K tokens of context. It became a default open-weight choice for developers in 2024 and spawned the Qwen2-Math, Qwen2-Audio, and Qwen2-VL specialist lines.

The licensing shift was the release's most consequential detail: every size except the 72B moved to Apache 2.0, making Qwen2 genuinely open for commercial use at most scales. The 72B keeps a custom license with commercial-use conditions. Note that this entry covers the base Qwen2 family release; the later Qwen2.5 generation and the Qwen2-VL, Qwen2-Audio, and Qwen2-Math specialists are cataloged separately.