QbitAI· 量子位的朋友们·· 6 d agoPickAI score42
Alibaba releases Qwen4-Max with upgraded agent and programming capabilities, 58% cost reduction
阿里正式发布Qwen4,智能体及编程能力进阶,迈向真实全场景
AI summary
Alibaba released the Qwen4-Max foundation model on November 16 with 2.4T total parameters and 71B activated, supporting million-token context. It ranks 6th globally in CodeArena front-end programming blind tests and scores 46.7 on Agents' Last Exam; compared to Qwen3.8-Max, professional coding evaluation token usage dropped nearly 50% and average cost per task completion decreased 58%.
Why it matters
Qwen4-Max's concrete progress in programming and agent capabilities, cost reduction ratios, and real-world applications (materials discovery, automotive manufacturing, financial research) are all quantitatively supported, making it a valuable benchmark.
Source: QbitAI · qbitai.com