Claude Sonnet 4.6 · BrowseComp(1,266 题;单代理与多代理按协议区分) 74.72% · 主文证据(ev46-main-browsecomp)

Claude Sonnet 4.6 的 BrowseComp · 1,266 题;单代理与多代理按协议区分 官方出处。No thinking, max effort · 74.72%。见《Introducing Claude Sonnet 4.6》Evaluating Claude Sonnet 4.6, benchmark summary image, Agentic search row。证据 ev46-main-browsecomp。

原文与定位

Claude Sonnet 4.6 · BrowseComp(1,266 题;单代理与多代理按协议区分),No thinking, max effort配置,Accuracy 74.72%(具体成绩公布时间未确认)。见《Introducing Claude Sonnet 4.6》Evaluating Claude Sonnet 4.6, benchmark summary image, Agentic search row。 查看成绩条件

BrowseComp: 74.7%.
官方材料
Introducing Claude Sonnet 4.6 · 定位到章节
定位
主文 · Evaluating Claude Sonnet 4.6, benchmark summary image, Agentic search row
材料版本
Current page collected 2026-11-10; datePublished 2026-02-17; dateModified 2026-10-14T23:56:52Z
版本公布
未确认
披露时间
未确认 · 首发/后补未确认
采集时间
历史依据
暂无可靠历史版本依据
引用文本

Claude Sonnet 4.6 · BrowseComp(1,266 题;单代理与多代理按协议区分),No thinking, max effort配置,Accuracy 74.72%(具体成绩公布时间未确认)。官方出处 Introducing Claude Sonnet 4.6(主文,Current page collected 2026-11-10; datePublished 2026-02-17; dateModified 2026-10-14T23:56:52Z)Evaluating Claude Sonnet 4.6, benchmark summary image, Agentic search row,https://www.anthropic.com/news/claude-sonnet-4-6#evaluating-claude-sonnet-46。Benchlog https://benchlog.lekshome.cc/evidence/claude-sonnet-4-6/ev46-main-browsecomp/

关联成绩与条件

Claude Sonnet 4.6 · BrowseComp 成绩与配置
指标与成绩配置与评测条件证据关系
Accuracy74.72%原文 74.72% · 单位 %
公布 未确认

已更正 · 74.01%。System Card changelog dated 2026-03-06 states that an improved cheating-detection pipeline flagged nine additional unintended solutions; the best-scoring single-agent configuration was re-verified, not rerun.

No thinking, max effort
设置
Thinking disabled; max effort.
未知项
Sampling settings are not specified in every evaluation.
工具
工具条件未记录
推理预算
max
框架
未记录
协议
BrowseComp single-agent best configuration, 1,266 questions.
条件
Historical result recorded in the current System Card changelog before the 2026-03-06 correction. Main article's summary image still shows the rounded 74.7%. No rerun was performed for the single-agent correction.

直接支持该成绩