Qwen3 14B (Reasoning)

Qwen3 14B (Reasoning) is a dense large language model developed by Alibaba Cloud's Qwen team, released in April 2025 as part of the Qwen3 model family. The model contains 14.8 billion parameters and is built on a causal transformer architecture with 40 layers and Grouped Query Attention. A defining feature of this model is its hybrid reasoning capability, which allows it to switch between a specialized 'thinking' mode for multi-step logical deduction and a 'non-thinking' mode for general-purpose conversation. Trained on a corpus of 36 trillion tokens, the model supports 119 languages and exhibits strong performance in mathematics, programming, and scientific reasoning. It natively supports context lengths of 32,768 tokens, extendable to 131,072 tokens using YaRN scaling, and is released under an Apache 2.0 license.