Alibaba's open-weight LLM family that rivals the best closed models - and runs on your own hardware.
Qwen (pronounced 'kwen') is the large language model family from Alibaba's Tongyi Lab. The 2.5 generation spans from a 0.5B edge model to a 72B flagship that beats many closed models on coding and math benchmarks. Because the weights are open, you can run Qwen in your own VPC, fine-tune it, and avoid sending data to a third party. It also ships with strong vision, audio, and coding variants.
Who it's for: Engineers and teams who want frontier-level quality with data sovereignty, plus hobbyists running models on consumer GPUs.
Apache 2.0 licensed models from 0.5B to 72B - download and deploy anywhere.
Qwen-Coder variants score near the top of HumanEval and LiveCodeBench.
Vision (VL), audio, and math-specialized variants in one family.
Strong in 30+ languages, with exceptional Chinese-English balance.
For teams that need a powerful model without sending data off-site, Qwen 2.5 is the best open option in 2026. Pair it with Ollama or vLLM and you have a private GPT-class assistant.