返回首页
RadarAI··论文与技术

Alibaba Qwen4-Max Launch with Performance Contradictions and Cost Reductions

中文摘要

阿里发布Qwen4-Max,基准与代理表现存矛盾。该模型采用MoE架构及隐式缓存,将运营成本降低了60%以上。

English Summary

Alibaba launched Qwen4-Max with performance contradictions between benchmarks and agents, leveraging MoE architecture and implicit caching to reduce costs by over 60%.

原文节选

📌 一句话摘要 Qwen4-Max is released showing conflicting benchmark results—ranking 43rd in test scores but 14th in real agent tasks—and features a new MoE architecture with over 60% cost reduction through implicit caching. 📝 详细摘要 ...