Ethan Mollick 转述 Artificial Analysis 新基准：开源权重模型落后幅度扩大
Artificial Analysis finally moves to closed benchmarks (goodbye GDPval-AA), and finds that open weights models are falling further behind than they had previou
Ethan Mollick 引用 Artificial Analysis 的 v5 基准结果指出，开源权重模型落后的幅度比此前测得的更大，并称这与业界尚未出现 Fable 级开源模型的普遍感受一致。Artificial Analysis 的数据显示，榜首 Kimi K3.1 Pro（权重尚未发布）得分 46，落后领先者 12 分，而 v4.3 时差距为 3 分；已发布权重的模型中 DeepSeek V4.1 Flash 1017（max）和 GLM-5.3（max）最高为 32 分，Claude Haiku 5.5（max）得分高于它们。差距在 HLE-Diamond 和 Terminal-Bench-Science 上最大，Kimi K3.1 Pro 分别为 37% 和 34%，Claude Opus 5.5（max）为 68%，GPT-6 Astra（max）为 63%。
查看原帖在 v5 中，开放权重模型进一步落后。领先的 Kimi K3.1 Pro（权重尚未发布）得分为 46，落后领先者 12 分，而 v4.3 时的差距为 3 分。在已发布权重的模型中，DeepSeek V4.1 Flash 1017（max）和 GLM-5.3（max）得分最高，为 32 分，而 Claude Haiku 5.5（max）的得分高于它们。 差距在 HLE-Diamond 和 Terminal-Bench-Science 上最大，Kimi K3.1 Pro 的得分分别为 37% 和 34%，而 Claude Opus 5.5（max）为 68%，GPT-6 Astra（max）为 63%。
查看原文
Open weights models fall further behind in v5. The leader, Kimi K3.1 Pro (weights not yet released), scores 46, 12 points behind the leader, compared with a 3-point gap in v4.3. Of models with released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) score highest at 32, and Claude Haiku 5.5 (max) outscores all of them. The gap is widest on HLE-Diamond and Terminal-Bench-Science, where Kimi K3.1 Pro scores 37% and 34%, compared with 68% for Claude Opus 5.5 (max) and 63% for GPT-6 Astra (max).
正文
Artificial Analysis finally moves to closed benchmarks (goodbye GDPval-AA), and finds that open weights models are falling further behind than they had previously measured. Take with a grain of salt, but matches the general vibes that we haven't seen a Fable-class open model.