Alibabas new Qwen reasoning AI model sets open-source records

Alibaba's Qwen team has unveiled Qwen3-235B-A22B-Thinking-2507, an open-source reasoning AI model that significantly advances capabilities in logical reasoning, complex mathematics, science problems, and advanced coding. Leveraging a Mixture-of-Experts (MoE) architecture, the model activates only about 22 billion parameters out of its total 235 billion, enabling efficient processing while maintaining high performance, and features an unprecedented native context length of 262,144 tokens, facilitating the handling of extensive information for complex tasks. The model has achieved notable benchmarks, including a 92