PANews reported on September 24th that according to the Qwen official account, Alibaba Cloud launched Qwen3-Max , its largest model to date, with over 1 trillion parameters and 36 trillion tokens pre-trained . The preview version of Qwen3-Max-Instruct ranked third globally on the LMArena text rankings. The official version has improved performance in code generation and agent tasks, achieving a SWE-Bench Verified score of 69.6 and a Tau2-Bench score of 74.8 . The reasoning-enhanced version , Qwen3-Max-Thinking , achieved full scores on the AIME 25 and HMMT mathematical reasoning tests. The API is now open and compatible with OpenAI.PANews reported on September 24th that according to the Qwen official account, Alibaba Cloud launched Qwen3-Max , its largest model to date, with over 1 trillion parameters and 36 trillion tokens pre-trained . The preview version of Qwen3-Max-Instruct ranked third globally on the LMArena text rankings. The official version has improved performance in code generation and agent tasks, achieving a SWE-Bench Verified score of 69.6 and a Tau2-Bench score of 74.8 . The reasoning-enhanced version , Qwen3-Max-Thinking , achieved full scores on the AIME 25 and HMMT mathematical reasoning tests. The API is now open and compatible with OpenAI.