What is Qwen (Alibaba's LLM)?

Qwen is Alibaba's family of large language models, and it has long been one of the most downloaded open-weight model lines in the world. If you have browsed fine-tuned models on Hugging Face, odds are good many were built on a Qwen base.

The family is broad.

• chat models
• coding models
• vision-language models
• sizes from laptop-friendly to data-center-scale

The open releases ship under the permissive Apache 2.0 license, which is why startups and hobbyists build on them so freely.

Here is where things stood as of July 2026. Qwen3.5, released in February 2026 with open weights, is a mixture-of-experts model with 397 billion total parameters (17 billion active) and a 262K-token context window. April's Qwen3.6 line added smaller open models like the dense Qwen3.6-27B, still under Apache 2.0.

The plot twist: Alibaba's newest flagships, Qwen3.7-Max and Qwen3.7-Plus (May 2026), are API-only with no open weights. The strategy has split — mid-size models stay open to keep the ecosystem loyal, while frontier models go closed to make money. Whether that balance holds is one of the livelier questions in the open-source LLM world; you can see how current Qwen models stack up in the model tracker.