Humanity's Last Exam · 2,500-question benchmark · 测试定义与成绩

Humanity's Last Exam · 2,500-question benchmark 收录 22 次模型发布的评测披露或成绩。

测试版本与指标

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

  • Accuracy · % · 越高越好

各系列最新模型 · 公布成绩

常规展示范围。排名按指标与评测协议分组,已记录的工具、预算及框架分别展示。条件未说明的组保留公布值供查证。

Accuracy %

禁止工具 / max / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget; no context compaction.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / max / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget; no context compaction.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5Max45.9%

允许工具 / low / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / low / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5Low38.6%

允许工具 / medium / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / medium / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5Medium45%

允许工具 / high / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / high / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5High50.1%

允许工具 / xhigh / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / xhigh / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5Xhigh54.8%

允许工具 / max / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / max / 框架未记录 · Humanity’s Last Exam; with web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5Max57.4%

禁止工具 / low / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / low / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5Low30.9%

禁止工具 / medium / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / medium / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5Medium35.5%

禁止工具 / high / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / high / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5High39.7%

禁止工具 / xhigh / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / xhigh / 框架未记录 · Humanity’s Last Exam; no tools; 980k-token task budget.
模型与配置成绩(%)记录与证据
Claude Haiku 5.5Xhigh44.1%

允许工具 / max / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / max / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5Max64.5%

允许工具 / low / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / low / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5Low47.6%

允许工具 / medium / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / medium / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5Medium50.5%

允许工具 / high / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / high / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5High56.7%

允许工具 / xhigh / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / xhigh / 框架未记录 · Humanity's Last Exam, 2,500 questions; tool-assisted configuration.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5Xhigh62%

禁止工具 / low / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / low / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5Low41.4%

禁止工具 / medium / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / medium / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5Medium43.2%

禁止工具 / high / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / high / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5High47.9%

禁止工具 / xhigh / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / xhigh / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5Xhigh53%

禁止工具 / max / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / max / 框架未记录 · Humanity's Last Exam, 2,500 questions; reasoning-only configuration without tools.
模型与配置成绩(%)记录与证据
Claude Sonnet 5.5Max56.9%

允许工具 / max / 框架未记录 · Humanity's Last Exam; with tools。

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / max / 框架未记录 · Humanity's Last Exam; with tools。
模型与配置成绩(%)记录与证据
Claude Opus 5.5Max67.7%

禁止工具 / low / 框架未记录 · HLE; no tools; low effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / low / 框架未记录 · HLE; no tools; low effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.
模型与配置成绩(%)记录与证据
Claude Opus 5.5Low52.9%

禁止工具 / medium / 框架未记录 · HLE; no tools; medium effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / medium / 框架未记录 · HLE; no tools; medium effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.
模型与配置成绩(%)记录与证据
Claude Opus 5.5Medium59%

禁止工具 / high / 框架未记录 · HLE; no tools; high effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / high / 框架未记录 · HLE; no tools; high effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.
模型与配置成绩(%)记录与证据
Claude Opus 5.5High59.5%

禁止工具 / xhigh / 框架未记录 · HLE; no tools; xhigh effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / xhigh / 框架未记录 · HLE; no tools; xhigh effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.
模型与配置成绩(%)记录与证据
Claude Opus 5.5Xhigh62.8%

禁止工具 / max / 框架未记录 · HLE; no tools; max effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / max / 框架未记录 · HLE; no tools; max effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.
模型与配置成绩(%)记录与证据
Claude Opus 5.5Max64.4%

允许工具 / low / 框架未记录 · HLE; web search, web fetch, programmatic tools, and code execution; low effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / low / 框架未记录 · HLE; web search, web fetch, programmatic tools, and code execution; low effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.
模型与配置成绩(%)记录与证据
Claude Opus 5.5Low57.4%

允许工具 / medium / 框架未记录 · HLE; web search, web fetch, programmatic tools, and code execution; medium effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / medium / 框架未记录 · HLE; web search, web fetch, programmatic tools, and code execution; medium effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.
模型与配置成绩(%)记录与证据
Claude Opus 5.5Medium63%

允许工具 / high / 框架未记录 · HLE; web search, web fetch, programmatic tools, and code execution; high effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / high / 框架未记录 · HLE; web search, web fetch, programmatic tools, and code execution; high effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.
模型与配置成绩(%)记录与证据
Claude Opus 5.5High63.9%

允许工具 / xhigh / 框架未记录 · HLE; web search, web fetch, programmatic tools, and code execution; xhigh effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / xhigh / 框架未记录 · HLE; web search, web fetch, programmatic tools, and code execution; xhigh effort; adaptive thinking; 1M total-token cap; no context compaction; Claude Opus 4.6 grader.
模型与配置成绩(%)记录与证据
Claude Opus 5.5Xhigh66.4%

禁止工具 / xhigh / 框架未记录 · 条件未说明

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / xhigh / 框架未记录 · 条件未说明
模型与配置成绩(%)记录与证据
GPT-5.5Xhigh41.4%

禁止工具 / 推理预算未记录 / 框架未记录 · 条件未说明

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / 推理预算未记录 / 框架未记录 · 条件未说明
模型与配置成绩(%)记录与证据
GPT-5.5Pro43.1%

允许工具 / xhigh / 框架未记录 · Humanity's Last Exam; with tools.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / xhigh / 框架未记录 · Humanity's Last Exam; with tools.
名次模型与配置成绩(%)记录与证据
1GPT-5.4 miniXhigh41.5%
2GPT-5.4 nanoXhigh37.7%

禁止工具 / xhigh / 框架未记录 · Humanity's Last Exam; without tools.

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / xhigh / 框架未记录 · Humanity's Last Exam; without tools.
名次模型与配置成绩(%)记录与证据
1GPT-5.4 miniXhigh28.2%
2GPT-5.4 nanoXhigh24.3%

禁止工具 / high / 框架未记录 · Full set (text + multimodal); no tools; Gemini 3.1 Pro Thinking (High).

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 禁止工具 / high / 框架未记录 · Full set (text + multimodal); no tools; Gemini 3.1 Pro Thinking (High).
模型与配置成绩(%)记录与证据
Gemini 3.1 ProHigh44.4%

允许工具 / high / 框架未记录 · Full set (text + multimodal); Search (blocklist) + Code; Gemini 3.1 Pro Thinking (High).

Humanity's Last Exam · 2,500-question benchmark · Accuracy · 允许工具 / high / 框架未记录 · Full set (text + multimodal); Search (blocklist) + Code; Gemini 3.1 Pro Thinking (High).
模型与配置成绩(%)记录与证据
Gemini 3.1 ProHigh51.4%

历史记录 · 22 次发布

  • Claude Haiku 5.5本次发布有 10 条成绩记录,含 10 条精确值(可含历史记录)
  • Claude Sonnet 5.5本次发布有 10 条成绩记录,含 10 条精确值(可含历史记录)
  • Claude Opus 5.5本次发布有 10 条成绩记录,含 10 条精确值(可含历史记录)
  • GPT-6 Astra本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)
  • Claude Opus 5本次发布有 12 条成绩记录,含 12 条精确值(可含历史记录)
  • Claude Sonnet 5本次发布有 6 条成绩记录,含 6 条精确值(可含历史记录)
  • Claude Opus 4.8本次发布有 2 条成绩记录,含 2 条精确值(可含历史记录)
  • GPT-5.5本次发布有 2 条成绩记录,含 2 条精确值(可含历史记录)
  • Claude Opus 4.7本次发布有 7 条成绩记录,含 7 条精确值(可含历史记录)
  • GPT-5.4 mini本次发布有 2 条成绩记录,含 2 条精确值(可含历史记录)
  • GPT-5.4 nano本次发布有 2 条成绩记录,含 2 条精确值(可含历史记录)
  • GPT-5.4本次发布有 4 条成绩记录,含 4 条精确值(可含历史记录)
  • Gemini 3.1 Pro本次发布有 2 条成绩记录,含 2 条精确值(可含历史记录)
  • Claude Sonnet 4.6本次发布有 2 条成绩记录,含 2 条精确值(可含历史记录)
  • Claude Opus 4.6本次发布有 2 条成绩记录,含 2 条精确值(可含历史记录)
  • GPT-5.2本次发布有 4 条成绩记录,含 4 条精确值(可含历史记录)
  • Claude Opus 4.5本次发布有 3 条成绩记录,含 3 条精确值(可含历史记录)
  • Gemini 3 Pro本次发布有 2 条成绩记录,含 2 条精确值(可含历史记录)
  • GPT-5本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)
  • GPT-5 mini本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)
  • GPT-5 nano本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)
  • Gemini 2.5 Pro (GA)本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)