Claude Sonnet 5.5 · WANDR · 500-task wide research benchmark · 成绩与配置

Claude Sonnet 5.5 的 WANDR · 500-task wide research benchmark 记录。本次发布有 5 条成绩记录,含 5 条精确值(可含历史记录),附 2 条来源定位。

成绩与评测条件

Claude Sonnet 5.5 · WANDR 成绩与配置
指标与成绩配置与评测条件出处
Soft F110%原文 10.0% · 单位 %
公布 未确认
Low effort
设置
Low reasoning effort.
未知项
Sampling settings not stated unless recorded per result.
工具
允许工具
推理预算
low
框架
未记录
协议
WANDR, 500 tasks; soft F1 averaged across tasks.
条件
Figure labels the effort Low; offline web search and web fetch over a frozen index; Claude Opus 4.6 grader; 980k-token task budget. The report says the modified tools and grading differ from Perplexity's harness.
在排名中查看此成绩
Soft F129.9%原文 29.9% · 单位 %
公布 未确认
Medium effort
设置
Medium reasoning effort.
未知项
Sampling settings not stated unless recorded per result.
工具
允许工具
推理预算
med
框架
未记录
协议
WANDR, 500 tasks; soft F1 averaged across tasks.
条件
Figure labels the effort Med; offline web search and web fetch over a frozen index; Claude Opus 4.6 grader; 980k-token task budget. The report says the modified tools and grading differ from Perplexity's harness.
在排名中查看此成绩
Soft F156.3%原文 56.3% · 单位 %
公布 未确认
High effort
设置
High reasoning effort.
未知项
Sampling settings not stated unless recorded per result.
工具
允许工具
推理预算
high
框架
未记录
协议
WANDR, 500 tasks; soft F1 averaged across tasks.
条件
Figure labels the effort High; offline web search and web fetch over a frozen index; Claude Opus 4.6 grader; 980k-token task budget. The report says the modified tools and grading differ from Perplexity's harness.
在排名中查看此成绩
Soft F166.6%原文 66.6% · 单位 %
公布 未确认
Xhigh effort
设置
Xhigh reasoning effort.
未知项
Sampling settings not stated unless recorded per result.
工具
允许工具
推理预算
xhigh
框架
未记录
协议
WANDR, 500 tasks; soft F1 averaged across tasks.
条件
Figure labels the effort Xhigh; offline web search and web fetch over a frozen index; Claude Opus 4.6 grader; 980k-token task budget. The report says the modified tools and grading differ from Perplexity's harness.
在排名中查看此成绩
Soft F170%原文 70.0% · 单位 %
公布 未确认
Max effort
设置
Max reasoning effort.
未知项
Adaptive thinking status is benchmark-specific.
工具
允许工具
推理预算
max
框架
未记录
协议
WANDR, 500 tasks; soft F1 averaged across tasks.
条件
Figure labels the effort Max; offline web search and web fetch over a frozen index; Claude Opus 4.6 grader; 980k-token task budget. The report says the modified tools and grading differ from Perplexity's harness.
在排名中查看此成绩

出处与披露

测试版本与指标

别名 Wide ANd Deep Research

  • Soft F1 · % · 越高越好