Qwen 2.5 / Qwen 3

Qwen is Alibaba Cloud's open-weight large language model family. Qwen 2.5 spans dense models from 0.5B to 72B parameters with strong multilingual coverage and capable coding and math performance. Qwen 3 continues the line with dense variants from 0.6B to 32B plus Mixture-of-Experts models at 30B-A3B and 235B-A22B, and introduces hybrid reasoning: the models switch between a thinking mode that produces extended reasoning chains for hard math, coding, and science problems and a non-thinking mode for fast general-purpose chat. Recent releases support context windows of 256K tokens, expandable to 1 million, cover more than 100 languages, and handle tool use and agent workflows. Weights are published on Hugging Face and ModelScope under the Apache 2.0 license, making Qwen a common base for community fine-tunes. Researchers, developers, and enterprises use the models for multilingual assistants, coding tools, and reasoning-heavy applications.