Claude Opus 5.5 · WANDR(500-task wide research benchmark) · 附件证据(ev-op55-card-wandr)

Claude Opus 5.5 的 WANDR · 500-task wide research benchmark 官方出处。见《Claude Opus 5.5 System Card》§8.11.3, pp.188-189, Figure 8.11.3.A。证据 ev-op55-card-wandr。

原文与定位

Claude Opus 5.5 · WANDR(500-task wide research benchmark),WANDR low effort, frozen-index harness配置,Soft F1 31.2%(具体成绩公布时间未确认)。见《Claude Opus 5.5 System Card》§8.11.3, pp.188-189, Figure 8.11.3.A。 查看成绩条件

Claude Opus 5.5 · WANDR(500-task wide research benchmark),WANDR medium effort, frozen-index harness配置,Soft F1 62.8%(具体成绩公布时间未确认)。见《Claude Opus 5.5 System Card》§8.11.3, pp.188-189, Figure 8.11.3.A。 查看成绩条件

Claude Opus 5.5 · WANDR(500-task wide research benchmark),WANDR high effort, frozen-index harness配置,Soft F1 67.3%(具体成绩公布时间未确认)。见《Claude Opus 5.5 System Card》§8.11.3, pp.188-189, Figure 8.11.3.A。 查看成绩条件

Claude Opus 5.5 · WANDR(500-task wide research benchmark),WANDR xhigh effort, frozen-index harness配置,Soft F1 71.3%(具体成绩公布时间未确认)。见《Claude Opus 5.5 System Card》§8.11.3, pp.188-189, Figure 8.11.3.A。 查看成绩条件

Claude Opus 5.5 · WANDR(500-task wide research benchmark),WANDR max effort, frozen-index harness配置,Soft F1 72.3%(具体成绩公布时间未确认)。见《Claude Opus 5.5 System Card》§8.11.3, pp.188-189, Figure 8.11.3.A。 查看成绩条件

WANDR soft F1 score versus average cost per task across reasoning effort levels at a 980k token task budget.
官方材料
Claude Opus 5.5 System Card · PDF 第 188 页
定位
附件 · §8.11.3, pp.188-189, Figure 8.11.3.A
材料版本
官方当前 System Card;封面日期 2026-09-22;本地 PDF/TXT:本地留存副本、current-opus-5-5-system-card.txt;230 页全文核读;当前修订时间未知
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 Opus 5.5 · WANDR(500-task wide research benchmark),WANDR low effort, frozen-index harness配置,Soft F1 31.2%(具体成绩公布时间未确认);Claude Opus 5.5 · WANDR(500-task wide research benchmark),WANDR medium effort, frozen-index harness配置,Soft F1 62.8%(具体成绩公布时间未确认);Claude Opus 5.5 · WANDR(500-task wide research benchmark),WANDR high effort, frozen-index harness配置,Soft F1 67.3%(具体成绩公布时间未确认);Claude Opus 5.5 · WANDR(500-task wide research benchmark),WANDR xhigh effort, frozen-index harness配置,Soft F1 71.3%(具体成绩公布时间未确认);Claude Opus 5.5 · WANDR(500-task wide research benchmark),WANDR max effort, frozen-index harness配置,Soft F1 72.3%(具体成绩公布时间未确认)。官方出处 Claude Opus 5.5 System Card(附件,官方当前 System Card;封面日期 2026-09-22;本地 PDF/TXT:本地留存副本、current-opus-5-5-system-card.txt;230 页全文核读;当前修订时间未知)§8.11.3, pp.188-189, Figure 8.11.3.A,https://www.anthropic.com/claude-opus-5-5-system-card#page=188。Benchlog https://benchlog.lekshome.cc/evidence/claude-opus-5-5/ev-op55-card-wandr/

关联成绩与条件

Claude Opus 5.5 · WANDR 成绩与配置
指标与成绩配置与评测条件证据关系
Soft F131.2%原文 31.2% · 单位 %
公布 未确认
WANDR low effort, frozen-index harness
设置
Explicit low effort; web search and web fetch limited to a frozen index; programmatic tool calling and code execution; 980k-token task budget.
未知项
Model run date and direct API reproduction are not stated.
工具
允许工具
推理预算
low
框架
未记录
协议
WANDR; low effort; web search/fetch on frozen index, programmatic tools, code execution; 980k-token budget; five-hundred-task mean; Claude Opus 4.6 grader.
条件
Modified prompts and grading; not directly comparable to Perplexity/live-web competitor results. Cost is not inferred from the chart. Source-wide conditions: 500 tasks; frozen-index web search/fetch; Claude Opus 4.6 grader; modified prompts and scoring; not directly comparable to Perplexity/live-web competitor results.

直接支持该成绩

在排名中查看此成绩
Soft F162.8%原文 62.8% · 单位 %
公布 未确认
WANDR medium effort, frozen-index harness
设置
Explicit medium effort; web search and web fetch limited to a frozen index; programmatic tool calling and code execution; 980k-token task budget.
未知项
Model run date and direct API reproduction are not stated.
工具
允许工具
推理预算
medium
框架
未记录
协议
WANDR; medium effort; web search/fetch on frozen index, programmatic tools, code execution; 980k-token budget; five-hundred-task mean; Claude Opus 4.6 grader.
条件
Modified prompts and grading; not directly comparable to Perplexity/live-web competitor results. Cost is not inferred from the chart. Source-wide conditions: 500 tasks; frozen-index web search/fetch; Claude Opus 4.6 grader; modified prompts and scoring; not directly comparable to Perplexity/live-web competitor results.

直接支持该成绩

在排名中查看此成绩
Soft F167.3%原文 67.3% · 单位 %
公布 未确认
WANDR high effort, frozen-index harness
设置
Explicit high effort; web search and web fetch limited to a frozen index; programmatic tool calling and code execution; 980k-token task budget.
未知项
Model run date and direct API reproduction are not stated.
工具
允许工具
推理预算
high
框架
未记录
协议
WANDR; high effort; web search/fetch on frozen index, programmatic tools, code execution; 980k-token budget; five-hundred-task mean; Claude Opus 4.6 grader.
条件
Modified prompts and grading; not directly comparable to Perplexity/live-web competitor results. Cost is not inferred from the chart. Source-wide conditions: 500 tasks; frozen-index web search/fetch; Claude Opus 4.6 grader; modified prompts and scoring; not directly comparable to Perplexity/live-web competitor results.

直接支持该成绩

在排名中查看此成绩
Soft F171.3%原文 71.3% · 单位 %
公布 未确认
WANDR xhigh effort, frozen-index harness
设置
Explicit xhigh effort; web search and web fetch limited to a frozen index; programmatic tool calling and code execution; 980k-token task budget.
未知项
Model run date and direct API reproduction are not stated.
工具
允许工具
推理预算
xhigh
框架
未记录
协议
WANDR; xhigh effort; web search/fetch on frozen index, programmatic tools, code execution; 980k-token budget; five-hundred-task mean; Claude Opus 4.6 grader.
条件
Modified prompts and grading; not directly comparable to Perplexity/live-web competitor results. Cost is not inferred from the chart. Source-wide conditions: 500 tasks; frozen-index web search/fetch; Claude Opus 4.6 grader; modified prompts and scoring; not directly comparable to Perplexity/live-web competitor results.

直接支持该成绩

在排名中查看此成绩
Soft F172.3%原文 72.3% · 单位 %
公布 未确认
WANDR max effort, frozen-index harness
设置
Explicit max effort; web search and web fetch limited to a frozen index; programmatic tool calling and code execution; 980k-token task budget.
未知项
Model run date and direct API reproduction are not stated.
工具
允许工具
推理预算
max
框架
未记录
协议
WANDR; max effort; web search/fetch on frozen index, programmatic tools, code execution; 980k-token budget; five-hundred-task mean; Claude Opus 4.6 grader.
条件
Modified prompts and grading; not directly comparable to Perplexity/live-web competitor results. Cost is not inferred from the chart. Source-wide conditions: 500 tasks; frozen-index web search/fetch; Claude Opus 4.6 grader; modified prompts and scoring; not directly comparable to Perplexity/live-web competitor results.

直接支持该成绩

在排名中查看此成绩