Skip to content
中文

Trending

Past 48Most discussed in hours AI industry1 stories · Nov 17, 2026 22:30Updated
NO.01Comparison unavailable

阿里发布Qwen4-Max,仅托管提供

阿里巴巴发布Qwen4代首款模型Qwen4-Max,总参数2.4T, 每token激活71B,支持文本, 图像, 视频输入和1M token上下文。该模型仅通过QwenCloud API, Token Plan, Qwen Studio, QwenWork和Qoder等托管渠道提供,不开源权重,因此无法自行部署或使用量化版本;默认限速为每分钟500万token。 据量子位报道,Qwen4-Max在CodeArena前端编程盲测中位列全球第6,Agents' Last Exam(ALE)取得46.7分;缓存命中每百万token低至0.5元,完成单任务的平均费用较上代模型下降6成。SiliconANGLE称,该模型主干配有3280亿N-gram参数充当缓存,采用四通道残差结构并带门控,注意力机制引入Gated DeltaNet与QSA以降低处理长提示词的硬件需求和基础设施成本。 关于开源,量子位称Qwen4系列中的中小尺寸模型将陆续开源发布;SiliconANGLE称阿里计划后续开源一批基于相同技术的较小模型,并开源Qwen-Live-Delivery-Pl…

Latest developments阿里发布 Qwen4-Max,训练与注意力机制均有优化

SiliconANGLE · AI, MarkTechPost and 3 sources·3participants
15
Activity index
How is activity calculated? About this list

Activity counts independent accounts and organizations discussing the same event. Repeated posts count once. Recent discussion weighs more, halving every 24 hours. It measures attention, not reporting quality.

The list covers the past 48 hours. Trends compare the same continuously observed sources, not the whole internet. No trend is shown without comparable history.

Source lists include publicly readable reports. Other accounts and organizations may contribute to discussion without being listed. One organization across platforms counts once, so participant counts may be lower than source counts. Open a story to explore the coverage.

Surging
Discussion rising rapidly
New
First reported within 6 hours
Developing
Discussion still growing