About
Qwen3 is the third generation of Alibaba Group's large language model family, announced on 29 April 2025. The series includes dense models (0.6B, 1.7B, 4B, 8B, 14B, 32B) and Mixture-of-Experts models (30B-A3B and 235B-A22B). The key innovation is a hybrid operating mode: thinking mode — the model generates step-by-step reasoning inside a <think>…</think> block before the final answer — and non-thinking mode — a direct response analogous to previous Qwen2.5 models. Switching modes is done via the enable_thinking parameter or /think and /no_think flags in the prompt. Qwen3 models were trained on approximately 36 trillion tokens covering 119 languages and dialects (three times more than Qwen2.5). All models are publicly available under the Apache 2.0 licence.