3 6 Ke·2026-01-27 03:08Core Insights - Alibaba has launched Qwen3-Max-Thinking, its most advanced reasoning model, which competes effectively with top models like GPT-5.2-Thinking and Gemini 3 Pro across 19 benchmark tests [1][4]. Model Capabilities - Qwen3-Max-Thinking demonstrates strong performance in various benchmarks, achieving scores close to or exceeding those of leading models in categories such as Knowledge, STEM, Reasoning, and Tool Use [3]. - The model features adaptive tool invocation capabilities, allowing it to automatically call upon search engines and code interpreters as needed, enhancing user experience [4][10]. Technical Innovations - The model employs a unique reasoning approach that focuses on efficient use of computational resources, prioritizing intelligent reasoning processes over merely increasing parallel reasoning paths [4][16]. - Qwen3-Max-Thinking integrates a context window of 256k and is believed to have over 1 trillion parameters, similar to its preview version [4]. Pricing and Accessibility - The API for Qwen3-Max-Thinking is available at a competitive rate of 2.5 yuan per million input tokens and 10 yuan per million output tokens, providing good value for users [5][6]. Additional Features - On the same day, Alibaba also released the Qwen3-TTS series, which includes advanced voice synthesis capabilities such as voice cloning and natural language-based voice control [7]. - The model's ability to combine search and code interpretation has been tested, showing its capability to handle complex tasks effectively, although some limitations in search accuracy were noted [10][15]. Industry Context - Alibaba's advancements in AI models reflect a broader trend in the industry towards optimizing computational efficiency and enhancing reasoning capabilities, addressing the constraints of available computational power in AI research [26].