Claude Opus · DRACO · 100-task Perplexity deep-research benchmark · 历代记录

按发布时间查看 Claude Opus 的 4 次发布在 DRACO · 100-task Perplexity deep-research benchmark 的成绩、配置与披露。

仅为直接前代间的唯一、有效精确值计算差值,要求单位、指标与明确记录的配置和条件一致。

Claude Opus 4.7

· 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 4.7 · DRACO 成绩与配置
指标与成绩配置与评测条件出处
Normalized score77.7%原文 77.7% · 单位 %
公布 未确认
DRACO,max effort
设置
100 tasks; web search, web fetch, programmatic tool calling and code execution; adaptive thinking, max effort; 1M token limit; context compaction at 200K; five independent Opus 4.6 grading runs per response, mean normalized score.
未知项
未另记
工具
允许工具
推理预算
max
框架
未记录
协议
DRACO; normalized score over 100 curated research tasks.
条件
Adaptive thinking, max effort; web search, web fetch, programmatic tools and code execution; 1M token limit; compaction at 200K; single model run averaged over five independent Opus 4.6 grading runs.
在排名中查看此成绩

Claude Opus 4.8

· 本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)

与直接前代的记录缺少满足差值计算的同条件依据。

Claude Opus 4.8 · DRACO 成绩与配置
指标与成绩配置与评测条件出处
Normalized score80.4%原文 80.4% · 单位 %
公布 未确认
DRACO single-agent
设置
100 curated tasks;web search、web fetch、programmatic tool calling、code execution;adaptive max effort;1M token limit,200K 触发 context compaction;模型回答单次执行,Opus 4.6 judge 对每个回答评分 5 次取平均。
未知项
未另记
工具
允许工具
推理预算
max
框架
未记录
协议
DRACO, 100 curated tasks; single model response per task; five Opus 4.6 judging runs per response, averaged.
条件
Adaptive thinking/max effort; web search/fetch, programmatic tool calling and code execution; 1M token limit, context compaction triggered at 200K. Judge change from original benchmark limits direct comparison to paper headline values.
在排名中查看此成绩

Claude Opus 5

· 本次发布有 5 条成绩记录,含 5 条精确值(可含历史记录)

与直接前代的记录缺少满足差值计算的同条件依据。

Claude Opus 5 · DRACO 成绩与配置
指标与成绩配置与评测条件出处
Normalized score83.2%原文 83.2% · 单位 %
公布 未确认
DRACO · Low effort
设置
100 curated deep-research tasks; web search, web fetch, programmatic tool calling, and code execution; figure states a 980K-token budget; Opus 4.6 grades reports written to a file using the paper formula. Explicit adaptive-thinking effort: low.
未知项
The card does not state any additional sampling parameters beyond the cited evaluation description.
工具
允许工具
推理预算
low
框架
未记录
协议
DRACO; normalized score over 100 curated tasks; Opus 4.6 grader; report-file protocol.
条件
Explicit effort labels; figure budget 980K tokens; web search, fetch, programmatic tools and code execution; judge model and report-file scoring differ from original paper, so headline scores are not directly comparable.
在排名中查看此成绩
Normalized score86.8%原文 86.8% · 单位 %
公布 未确认
DRACO · Medium effort
设置
100 curated deep-research tasks; web search, web fetch, programmatic tool calling, and code execution; figure states a 980K-token budget; Opus 4.6 grades reports written to a file using the paper formula. Explicit adaptive-thinking effort: medium.
未知项
The card does not state any additional sampling parameters beyond the cited evaluation description.
工具
允许工具
推理预算
medium
框架
未记录
协议
DRACO; normalized score over 100 curated tasks; Opus 4.6 grader; report-file protocol.
条件
Explicit effort labels; figure budget 980K tokens; web search, fetch, programmatic tools and code execution; judge model and report-file scoring differ from original paper, so headline scores are not directly comparable.
在排名中查看此成绩
Normalized score87%原文 87% · 单位 %
公布 未确认
DRACO · High effort
设置
100 curated deep-research tasks; web search, web fetch, programmatic tool calling, and code execution; figure states a 980K-token budget; Opus 4.6 grades reports written to a file using the paper formula. Explicit adaptive-thinking effort: high.
未知项
The card does not state any additional sampling parameters beyond the cited evaluation description.
工具
允许工具
推理预算
high
框架
未记录
协议
DRACO; normalized score over 100 curated tasks; Opus 4.6 grader; report-file protocol.
条件
Explicit effort labels; figure budget 980K tokens; web search, fetch, programmatic tools and code execution; judge model and report-file scoring differ from original paper, so headline scores are not directly comparable.
在排名中查看此成绩
Normalized score88.1%原文 88.1% · 单位 %
公布 未确认
DRACO · Xhigh effort
设置
100 curated deep-research tasks; web search, web fetch, programmatic tool calling, and code execution; figure states a 980K-token budget; Opus 4.6 grades reports written to a file using the paper formula. Explicit adaptive-thinking effort: xhigh.
未知项
The card does not state any additional sampling parameters beyond the cited evaluation description.
工具
允许工具
推理预算
xhigh
框架
未记录
协议
DRACO; normalized score over 100 curated tasks; Opus 4.6 grader; report-file protocol.
条件
Explicit effort labels; figure budget 980K tokens; web search, fetch, programmatic tools and code execution; judge model and report-file scoring differ from original paper, so headline scores are not directly comparable.
在排名中查看此成绩
Normalized score88.6%原文 88.6% · 单位 %
公布 未确认
DRACO · Max effort
设置
100 curated deep-research tasks; web search, web fetch, programmatic tool calling, and code execution; figure states a 980K-token budget; Opus 4.6 grades reports written to a file using the paper formula. Explicit adaptive-thinking effort: max.
未知项
The card does not state any additional sampling parameters beyond the cited evaluation description.
工具
允许工具
推理预算
max
框架
未记录
协议
DRACO; normalized score over 100 curated tasks; Opus 4.6 grader; report-file protocol.
条件
Explicit effort labels; figure budget 980K tokens; web search, fetch, programmatic tools and code execution; judge model and report-file scoring differ from original paper, so headline scores are not directly comparable.
在排名中查看此成绩

Claude Opus 5.5

· 本次发布有 5 条成绩记录,含 5 条精确值(可含历史记录)

与直接前代的记录缺少满足差值计算的同条件依据。

Claude Opus 5.5 · DRACO 成绩与配置
指标与成绩配置与评测条件出处
Normalized score72.5%原文 72.5% · 单位 %
公布 未确认
DRACO low effort, in-house agent harness
设置
Explicit low effort; web search, web fetch, 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
DRACO; low effort; web search/fetch, programmatic tools, code execution; 980k-token budget; own agent harness; five judge grading runs per response, mean; Claude Opus 4.6 grader.
条件
Judge choice can shift absolute scores 10-25 points; not directly comparable to paper headline scores. Cost is shown only as a chart axis and is not added as an inferred exact value. Source-wide conditions: 100 tasks; web search, web fetch, programmatic tool calling, code execution; five judge grading runs per response, mean reported; Claude Opus 4.6 grader; not directly comparable to paper headline scores.
在排名中查看此成绩
Normalized score83.9%原文 83.9% · 单位 %
公布 未确认
DRACO medium effort, in-house agent harness
设置
Explicit medium effort; web search, web fetch, 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
DRACO; medium effort; web search/fetch, programmatic tools, code execution; 980k-token budget; own agent harness; five judge grading runs per response, mean; Claude Opus 4.6 grader.
条件
Judge choice can shift absolute scores 10-25 points; not directly comparable to paper headline scores. Cost is not inferred from the chart. Source-wide conditions: 100 tasks; web search, web fetch, programmatic tool calling, code execution; five judge grading runs per response, mean reported; Claude Opus 4.6 grader; not directly comparable to paper headline scores.
在排名中查看此成绩
Normalized score85%原文 85.0% · 单位 %
公布 未确认
DRACO high effort, in-house agent harness
设置
Explicit high effort; web search, web fetch, 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
DRACO; high effort; web search/fetch, programmatic tools, code execution; 980k-token budget; own agent harness; five judge grading runs per response, mean; Claude Opus 4.6 grader.
条件
Judge choice can shift absolute scores 10-25 points; not directly comparable to paper headline scores. Cost is not inferred from the chart. Source-wide conditions: 100 tasks; web search, web fetch, programmatic tool calling, code execution; five judge grading runs per response, mean reported; Claude Opus 4.6 grader; not directly comparable to paper headline scores.
在排名中查看此成绩
Normalized score86.7%原文 86.7% · 单位 %
公布 未确认
DRACO xhigh effort, in-house agent harness
设置
Explicit xhigh effort; web search, web fetch, 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
DRACO; xhigh effort; web search/fetch, programmatic tools, code execution; 980k-token budget; own agent harness; five judge grading runs per response, mean; Claude Opus 4.6 grader.
条件
Judge choice can shift absolute scores 10-25 points; not directly comparable to paper headline scores. Cost is not inferred from the chart. Source-wide conditions: 100 tasks; web search, web fetch, programmatic tool calling, code execution; five judge grading runs per response, mean reported; Claude Opus 4.6 grader; not directly comparable to paper headline scores.
在排名中查看此成绩
Normalized score87.4%原文 87.4% · 单位 %
公布 未确认
DRACO max effort, in-house agent harness
设置
Explicit max effort; web search, web fetch, 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
DRACO; max effort; web search/fetch, programmatic tools, code execution; 980k-token budget; own agent harness; five judge grading runs per response, mean; Claude Opus 4.6 grader.
条件
Judge choice can shift absolute scores 10-25 points; not directly comparable to paper headline scores. Cost is not inferred from the chart. Source-wide conditions: 100 tasks; web search, web fetch, programmatic tool calling, code execution; five judge grading runs per response, mean reported; Claude Opus 4.6 grader; not directly comparable to paper headline scores.
在排名中查看此成绩