Qwen is a family of large language models developed by Alibaba Cloud, the cloud computing division of Chinese conglomerate Alibaba Group. The first Qwen models were released in 2023, and the family has since expanded across multiple generations and specialized variants, including base language models, chat-tuned versions, vision-language models (Qwen-VL), coding-specialized models (Qwen-Coder), and audio models. Alibaba released most Qwen models under open-weight licenses, primarily Apache 2.0, which contributed to the family becoming, by 2025, one of the most downloaded model families on Hugging Face, both directly and through the large number of derivative models fine-tuned from Qwen checkpoints by the broader open-source community.
Model lineage
Qwen1 launched in 2023 at parameter scales up to 72 billion. Qwen2 followed in 2024 with improved multilingual coverage and longer context windows. Qwen2.5, released in September 2024, spanned a wide range of sizes from under a billion parameters up to 72 billion, alongside specialized coding and math variants, and was widely benchmarked as competitive with similarly sized open models from Meta AI's Llama family. Qwen3, released in 2025, introduced hybrid reasoning capability, letting a single model switch between a fast direct-answer mode and an extended Chain-of-thought reasoning mode, echoing a design direction also pursued by other labs racing to combine general chat capability with reasoning-model performance. Qwen3 was released in a range of sizes including Mixture of experts variants aimed at balancing capability against inference cost.
Position in the open-weights ecosystem
Qwen models are frequently used as a base for Fine-tuning by researchers, startups, and hobbyists, both because of their permissive licensing and because their performance on standard benchmarks such as MMLU and coding evaluations like HumanEval compared favorably with other open options. DeepSeek's DeepSeek-R1 distilled reasoning models, for example, used Qwen checkpoints as one of the base models for distillation. Qwen's rapid iteration and broad size range, from models small enough to run on consumer hardware to variants competitive with larger closed models, positioned Alibaba as one of the most active contributors to the open-weights ecosystem alongside DeepSeek, Mistral AI, and Meta.
Reception and strategic context
Qwen's releases are often read in the context of a broader wave of capable open-weights models coming out of Chinese AI labs through 2024 and 2025, alongside DeepSeek and others, which analysts and Western policymakers cited as evidence that China's AI industry remained competitive despite US export controls on advanced GPUs. Alibaba positioned Qwen both as a research contribution and as the foundation for its own commercial cloud AI offerings, integrating Qwen models into Alibaba Cloud's enterprise products. Independent evaluators noted that, like other China-developed models, hosted consumer-facing Qwen products applied content restrictions aligned with Chinese regulatory requirements on sensitive political topics, though the openly released model weights themselves did not enforce this when run independently by third parties.