Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader) · 附件证据(ev-op46-card-webarena)
Claude Opus 4.6 的 WebArena · Official WebArena grader, modified fuzzy_match subgrader 官方出处。见《Claude Opus 4.6 System Card》PDF p.38, Tables 2.20.A and 2.20.B。证据 ev-op46-card-webarena。
Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Average@5 score 68%(具体成绩公布时间未确认)。见《Claude Opus 4.6 System Card》PDF p.38, Tables 2.20.A and 2.20.B。 查看成绩条件
Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@1 68%(具体成绩公布时间未确认)。见《Claude Opus 4.6 System Card》PDF p.38, Tables 2.20.A and 2.20.B。 查看成绩条件
Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@2 71.7%(具体成绩公布时间未确认)。见《Claude Opus 4.6 System Card》PDF p.38, Tables 2.20.A and 2.20.B。 查看成绩条件
Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@3 72.5%(具体成绩公布时间未确认)。见《Claude Opus 4.6 System Card》PDF p.38, Tables 2.20.A and 2.20.B。 查看成绩条件
Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@4 73.4%(具体成绩公布时间未确认)。见《Claude Opus 4.6 System Card》PDF p.38, Tables 2.20.A and 2.20.B。 查看成绩条件
Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@5 74%(具体成绩公布时间未确认)。见《Claude Opus 4.6 System Card》PDF p.38, Tables 2.20.A and 2.20.B。 查看成绩条件
WebArena Opus 4.6: Average@5 68.0%; Pass@1 68.0%, Pass@2 71.7%, Pass@3 72.5%, Pass@4 73.4%, Pass@5 74.0%. The official grader changed fuzzy_match from GPT-4 to Claude Sonnet 4.5 and used a rewritten judge prompt; Average@5 is five independent runs.
Current PDF/TXT; local source 本地留存副本 and 本地留存副本; 213 pages; changelog through 2026-03-06; full text and visuals reviewed.
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 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Average@5 score 68%(具体成绩公布时间未确认);Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@1 68%(具体成绩公布时间未确认);Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@2 71.7%(具体成绩公布时间未确认);Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@3 72.5%(具体成绩公布时间未确认);Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@4 73.4%(具体成绩公布时间未确认);Claude Opus 4.6 · WebArena(Official WebArena grader, modified fuzzy_match subgrader),WebArena · 单策略模型配置,Pass@5 74%(具体成绩公布时间未确认)。官方出处 Claude Opus 4.6 System Card(附件,Current PDF/TXT; local source 本地留存副本 and 本地留存副本; 213 pages; changelog through 2026-03-06; full text and visuals reviewed.)PDF p.38, Tables 2.20.A and 2.20.B,https://www.anthropic.com/claude-opus-4-6-system-card#page=38。Benchlog https://benchlog.lekshome.cc/evidence/claude-opus-4-6/ev-op46-card-webarena/
关联成绩与条件
Claude Opus 4.6 · WebArena 成绩与配置
指标与成绩
配置与评测条件
证据关系
Average@5 score68%原文 68.0% · 单位 %
公布 未确认
WebArena · 单策略模型
设置
Computer Use API;截图与 DOM 浏览器工具;通用提示;单策略模型;effort 未说明。
未知项
PDF 未报告 reasoning effort。
工具
允许工具
推理预算
未记录
框架
未记录
协议
WebArena;single policy model;Computer Use API;official grader with fuzzy_match subgrader changed to Claude Sonnet 4.5 and rewritten judge prompt。
条件
Average@5, five independent runs; effort not reported. Multi-agent reference systems are not directly comparable and are not included as Opus 4.6 results.。
Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader) · Attachment evidence (ev-op46-card-webarena)
Official source for WebArena · Official WebArena grader, modified fuzzy_match subgrader in Claude Opus 4.6. See “Claude Opus 4.6 System Card” at PDF p.38, Tables 2.20.A and 2.20.B. Evidence ev-op46-card-webarena.
Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Average@5 score 68% (exact score publication date unconfirmed). See “Claude Opus 4.6 System Card” at PDF p.38, Tables 2.20.A and 2.20.B. View score conditions
Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@1 68% (exact score publication date unconfirmed). See “Claude Opus 4.6 System Card” at PDF p.38, Tables 2.20.A and 2.20.B. View score conditions
Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@2 71.7% (exact score publication date unconfirmed). See “Claude Opus 4.6 System Card” at PDF p.38, Tables 2.20.A and 2.20.B. View score conditions
Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@3 72.5% (exact score publication date unconfirmed). See “Claude Opus 4.6 System Card” at PDF p.38, Tables 2.20.A and 2.20.B. View score conditions
Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@4 73.4% (exact score publication date unconfirmed). See “Claude Opus 4.6 System Card” at PDF p.38, Tables 2.20.A and 2.20.B. View score conditions
Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@5 74% (exact score publication date unconfirmed). See “Claude Opus 4.6 System Card” at PDF p.38, Tables 2.20.A and 2.20.B. View score conditions
WebArena Opus 4.6: Average@5 68.0%; Pass@1 68.0%, Pass@2 71.7%, Pass@3 72.5%, Pass@4 73.4%, Pass@5 74.0%. The official grader changed fuzzy_match from GPT-4 to Claude Sonnet 4.5 and used a rewritten judge prompt; Average@5 is five independent runs.
Current PDF/TXT; local source 本地留存副本 and 本地留存副本; 213 pages; changelog through 2026-03-06; full text and visuals reviewed.
Version published
Unconfirmed
Disclosure date
Unconfirmed · Launch or later disclosure unconfirmed
Collected
Historical basis
No reliable historical version evidence available
Citation text
Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Average@5 score 68% (exact score publication date unconfirmed); Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@1 68% (exact score publication date unconfirmed); Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@2 71.7% (exact score publication date unconfirmed); Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@3 72.5% (exact score publication date unconfirmed); Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@4 73.4% (exact score publication date unconfirmed); Claude Opus 4.6 · WebArena (Official WebArena grader, modified fuzzy_match subgrader), configuration WebArena · 单策略模型, Pass@5 74% (exact score publication date unconfirmed). Official source: Claude Opus 4.6 System Card (Attachment, Current PDF/TXT; local source 本地留存副本 and 本地留存副本; 213 pages; changelog through 2026-03-06; full text and visuals reviewed.), PDF p.38, Tables 2.20.A and 2.20.B, https://www.anthropic.com/claude-opus-4-6-system-card#page=38. Benchlog https://benchlog.lekshome.cc/evidence/claude-opus-4-6/ev-op46-card-webarena/?lang=en
Related scores and conditions
Claude Opus 4.6 · WebArena scores and configurations
Metric and score
Configuration and evaluation conditions
Evidence relation
Average@5 score68%Reported 68.0% · Unit %
Published Unconfirmed
WebArena · 单策略模型
Settings
Computer Use API;截图与 DOM 浏览器工具;通用提示;单策略模型;effort 未说明。
Unknowns
PDF 未报告 reasoning effort。
Tools
Tools allowed
Reasoning budget
not recorded
Framework
not recorded
Protocol
WebArena;single policy model;Computer Use API;official grader with fuzzy_match subgrader changed to Claude Sonnet 4.5 and rewritten judge prompt。
Conditions
Average@5, five independent runs; effort not reported. Multi-agent reference systems are not directly comparable and are not included as Opus 4.6 results.。