Qwen3 is the 2025 generation of Alibaba's Qwen language models, whose models can either reason step by step or reply quickly, in sizes from 0.6 to 235 billion parameters.
In plain words
Before Qwen3, you often needed one model for fast chat and another for slow, careful reasoning. Qwen3 puts both behaviours in one model: a thinking mode and a non-thinking mode. A thinking budget lets users limit how much reasoning it does. The family has ordinary dense models and mixture-of-experts models, supports 119 languages and dialects, and is released under Apache 2.0.
A quick example
A typical app could send a simple greeting in non-thinking mode for an instant reply, then switch the same model to thinking mode for a hard maths question.
Where you'll meet it
- Open-weight chat and reasoning assistants.
- Multilingual apps and agent tool use.
Related
Sources
- PaperQwen3 Technical Report, Qwen Team (arXiv) · read 29 Sept 2026
- DocsQwen3, QwenLM (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.