Qwen3 235B-A22B

Qwen3 is the latest evolution in the Qwen LLM series, featuring both dense and MoE models with major advancements in reasoning, agent capabilities, multilingual support, and instruction following. It uniquely allows seamless switching between “thinking” (for complex logic, math, coding) and “non-thinking” modes (for fast, general dialogue), delivering strong performance across tasks. Qwen3 outperforms previous Qwen models in math, code, and logical reasoning, while also offering superior human alignment for creative writing, roleplay, and multi-turn conversations. It supports over 100 languages and excels at tool integration for agent-based tasks. The flagship model, Qwen3-235B-A22B, has 235B parameters (22B active), 94 layers, and a native context length of 32K, extendable to 131K with YaRN.