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
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
条件
Figure labels the effort Low; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.
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
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
条件
Figure labels the effort Med; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.
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
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
条件
Figure labels the effort High; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.
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
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
条件
Figure labels the effort Xhigh; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
条件
Figure labels the effort Max; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.
Official source for DRACO · 100-task Perplexity deep-research benchmark in Claude Sonnet 5.5. See “Claude Sonnet 5.5 System Card” at §8.11.2, Figure 8.11.2.A, PDF p.121; chart-label transcription. Evidence ev55-visual-draco-effort.
Attachment · §8.11.2, Figure 8.11.2.A, PDF p.121; chart-label transcription
Material version
PDF cover dated 2026-09-28; 148 pages; collected 2026-11-10
Version published
Unconfirmed
Disclosure date
Unconfirmed · Launch or later disclosure unconfirmed
Collected
Historical basis
No reliable historical version evidence available
Citation text
Claude Sonnet 5.5 · DRACO (100-task Perplexity deep-research benchmark), configuration Low effort, Normalized score 64.6% (exact score publication date unconfirmed); Claude Sonnet 5.5 · DRACO (100-task Perplexity deep-research benchmark), configuration Medium effort, Normalized score 71.2% (exact score publication date unconfirmed); Claude Sonnet 5.5 · DRACO (100-task Perplexity deep-research benchmark), configuration High effort, Normalized score 81% (exact score publication date unconfirmed); Claude Sonnet 5.5 · DRACO (100-task Perplexity deep-research benchmark), configuration Xhigh effort, Normalized score 85.7% (exact score publication date unconfirmed); Claude Sonnet 5.5 · DRACO (100-task Perplexity deep-research benchmark), configuration Max effort, Normalized score 87% (exact score publication date unconfirmed). Official source: Claude Sonnet 5.5 System Card (Attachment, PDF cover dated 2026-09-28; 148 pages; collected 2026-11-10), §8.11.2, Figure 8.11.2.A, PDF p.121; chart-label transcription, https://www.anthropic.com/claude-sonnet-5-5-system-card#page=121. Benchlog https://benchlog.lekshome.cc/evidence/claude-sonnet-5-5/ev55-visual-draco-effort/?lang=en
Related scores and conditions
Claude Sonnet 5.5 · DRACO scores and configurations
Metric and score
Configuration and evaluation conditions
Evidence relation
Normalized score64.6%Reported 64.6% · Unit %
Published Unconfirmed
Low effort
Settings
Low reasoning effort.
Unknowns
Sampling settings not stated unless recorded per result.
Tools
Tools allowed
Reasoning budget
low
Framework
not recorded
Protocol
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
Conditions
Figure labels the effort Low; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.
Sampling settings not stated unless recorded per result.
Tools
Tools allowed
Reasoning budget
med
Framework
not recorded
Protocol
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
Conditions
Figure labels the effort Med; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.
Sampling settings not stated unless recorded per result.
Tools
Tools allowed
Reasoning budget
high
Framework
not recorded
Protocol
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
Conditions
Figure labels the effort High; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.
Sampling settings not stated unless recorded per result.
Tools
Tools allowed
Reasoning budget
xhigh
Framework
not recorded
Protocol
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
Conditions
Figure labels the effort Xhigh; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.
DRACO, 100 curated tasks; normalized score from five independent rubric-grading runs per response.
Conditions
Figure labels the effort Max; web search, web fetch, programmatic tool calling, and code execution; 980k-token task budget; Claude Opus 4.6 grader. The report says its modified tools, prompts, and grader make results not directly comparable to Perplexity's version.