Claude Haiku 5.5 · Humanity's Last Exam · 2,500-question benchmark · 成绩与配置

Claude Haiku 5.5 的 Humanity's Last Exam · 2,500-question benchmark 记录。本次发布有 10 条成绩记录,含 10 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 Haiku 5.5 · Humanity's Last Exam 成绩与配置
指标与成绩配置与评测条件出处
Accuracy45.9%原文 45.9% no tools · 单位 %
公布 未确认
HLE without tools / max
设置
Thinking auto; max effort; no tools; 980k-token task budget; no context compaction.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
禁止工具
推理预算
max
框架
未记录
协议
Humanity’s Last Exam; no tools; 980k-token task budget; no context compaction.
条件
Max effort; this is the no-tools main-table result. The card’s effort figure also labels the same 45.9% max point.
在排名中查看此成绩
Accuracy38.6%原文 38.6% · 单位 %
公布 未确认
HLE with tools / low
设置
Thinking auto; low effort; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; no context compaction; Opus 4.6 grader.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
允许工具
推理预算
low
框架
未记录
协议
Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
条件
Low effort; no context compaction; Opus 4.6 grader; point read from the labeled figure.
在排名中查看此成绩
Accuracy45%原文 45.0% · 单位 %
公布 未确认
HLE with tools / medium
设置
Thinking auto; medium effort; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; no context compaction; Opus 4.6 grader.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
允许工具
推理预算
medium
框架
未记录
协议
Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
条件
Medium effort; no context compaction; Opus 4.6 grader; point read from the labeled figure.
在排名中查看此成绩
Accuracy50.1%原文 50.1% · 单位 %
公布 未确认
HLE with tools / high
设置
Thinking auto; high effort; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; no context compaction; Opus 4.6 grader.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
允许工具
推理预算
high
框架
未记录
协议
Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
条件
High effort; no context compaction; Opus 4.6 grader; point read from the labeled figure.
在排名中查看此成绩
Accuracy54.8%原文 54.8% · 单位 %
公布 未确认
HLE with tools / xhigh
设置
Thinking auto; xhigh effort; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; no context compaction; Opus 4.6 grader.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
允许工具
推理预算
xhigh
框架
未记录
协议
Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
条件
Xhigh effort; no context compaction; Opus 4.6 grader; point read from the labeled figure.
在排名中查看此成绩
Accuracy57.4%原文 57.4% · 单位 %
公布 未确认
HLE with tools / max
设置
Thinking auto; max effort; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; no context compaction; Opus 4.6 grader.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
允许工具
推理预算
max
框架
未记录
协议
Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
条件
Max effort; no context compaction; Opus 4.6 grader; point read from the labeled figure.
在排名中查看此成绩
Accuracy30.9%原文 30.9% · 单位 %
公布 未确认
HLE without tools / low
设置
Thinking auto; low effort; no tools; 980k-token task budget; no context compaction.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
禁止工具
推理预算
low
框架
未记录
协议
Humanity’s Last Exam; no tools; 980k-token task budget.
条件
Low effort; no context compaction; point read from the labeled figure.
在排名中查看此成绩
Accuracy35.5%原文 35.5% · 单位 %
公布 未确认
HLE without tools / medium
设置
Thinking auto; medium effort; no tools; 980k-token task budget; no context compaction.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
禁止工具
推理预算
medium
框架
未记录
协议
Humanity’s Last Exam; no tools; 980k-token task budget.
条件
Medium effort; no context compaction; point read from the labeled figure.
在排名中查看此成绩
Accuracy39.7%原文 39.7% · 单位 %
公布 未确认
HLE without tools / high
设置
Thinking auto; high effort; no tools; 980k-token task budget; no context compaction.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
禁止工具
推理预算
high
框架
未记录
协议
Humanity’s Last Exam; no tools; 980k-token task budget.
条件
High effort; no context compaction; point read from the labeled figure.
在排名中查看此成绩
Accuracy44.1%原文 44.1% · 单位 %
公布 未确认
HLE without tools / xhigh
设置
Thinking auto; xhigh effort; no tools; 980k-token task budget; no context compaction.
未知项
Per-effort trial count is not stated in the chart caption where applicable.
工具
禁止工具
推理预算
xhigh
框架
未记录
协议
Humanity’s Last Exam; no tools; 980k-token task budget.
条件
Xhigh effort; no context compaction; point read from the labeled figure.
在排名中查看此成绩

出处与披露

测试版本与指标

别名 HLE、Humanity’s Last Exam、Humanity's Last Exam

  • Accuracy · % · 越高越好