Welcome China Products Info!
loading
Release Info
    Categories
    China Products Info > 热点资讯 > China News >  Are China's Open-Source Models 'Exposed'? Artificial Analysis v5 Reshuffle Reorders the Leaderboard

    Are China's Open-Source Models 'Exposed'? Artificial Analysis v5 Reshuffle Reorders the Leaderboard

    时间:2026-10-30 11:42:23  编辑:graysky  来源:NetEase / Weijin Research, Oct 30, 2026  浏览:2次   【大】【中】【小】 网站投稿
    After Artificial Analysis upgraded its Intelligence Index to v5, Chinese open-source models collectively slipped as U.S. frontier closed models led. Kimi K3.1 Pro entered…

    After Artificial Analysis upgraded its Intelligence Index to v5, Chinese open-source models collectively slipped as U.S. frontier closed models led. Kimi K3.1 Pro entered the top ten but has not opened weights; among active open-weight domestic models, DeepSeek V4.1 Flash 1017 and GLM 5.3 ranked 14th and 15th, below Anthropic's entry-level Claude Haiku 5.5. The new index for the first time weights 60% on private test sets, adds FrontierCode 2.0 and Terminal-Bench-Science, and upgrades HLE, omniscience and physics reasoning. Some revived old critiques of distillation and benchmark overfitting, but analysts note the "ruler" changed: the v5 puts far more weight on real-world coding and long-horizon scientific workflows, not just general ability.

    Source: NetEase / Weijin Research, Oct 30, 2026 | 2026-10-30

    13775048177