1,235 evaluated tasks after 16 data-defect exclusions; adaptive thinking at max effort; public Messages API with production safeguards and fallback to Opus 4.8; internal reimplementation with bash and Python tools; five-run all-pass average; Sonnet 4.6 judge.
未知项
未另记
工具
允许工具
推理预算
max
框架
未记录
协议
Legal Agent Benchmark; full Anthropic evaluation set; all-pass rate.
条件
1,235 tasks after 16 data-defect exclusions; adaptive max effort; public Messages API safeguards/fallback; internal reduced-tool reimplementation; five-run average.
1,235 evaluated tasks after 16 data-defect exclusions; adaptive thinking at max effort; public Messages API with production safeguards and fallback to Opus 4.8; internal reimplementation with bash and Python tools; five-run all-pass average; Sonnet 4.6 judge.
未知项
未另记
工具
允许工具
推理预算
max
框架
未记录
协议
Legal Agent Benchmark; full Anthropic evaluation set; mean criterion-pass rate.
条件
1,235 tasks after 16 data-defect exclusions; adaptive max effort; public Messages API safeguards/fallback; internal reduced-tool reimplementation; five-run evaluation.
Claude Opus 5 · Legal Agent Benchmark · Public full set · Scores and configurations
Claude Opus 5 record for Legal Agent Benchmark · Public full set. 2 score records for this release, including 2 exact values (may include historical records). Includes 1 source location.
Claude Opus 5 · Legal Agent Benchmark scores and configurations
Metric and score
Configuration and evaluation conditions
Sources
All-pass rate23.58%Reported 23.58% ±0.48 (n=5) · Unit %
Published Unconfirmed
Legal Agent Benchmark · full set
Settings
1,235 evaluated tasks after 16 data-defect exclusions; adaptive thinking at max effort; public Messages API with production safeguards and fallback to Opus 4.8; internal reimplementation with bash and Python tools; five-run all-pass average; Sonnet 4.6 judge.
Unknowns
none recorded
Tools
Tools allowed
Reasoning budget
max
Framework
not recorded
Protocol
Legal Agent Benchmark; full Anthropic evaluation set; all-pass rate.
Conditions
1,235 tasks after 16 data-defect exclusions; adaptive max effort; public Messages API safeguards/fallback; internal reduced-tool reimplementation; five-run average.
Mean criterion-pass rate93.74%Reported 93.74% · Unit %
Published Unconfirmed
Legal Agent Benchmark · full set
Settings
1,235 evaluated tasks after 16 data-defect exclusions; adaptive thinking at max effort; public Messages API with production safeguards and fallback to Opus 4.8; internal reimplementation with bash and Python tools; five-run all-pass average; Sonnet 4.6 judge.
Unknowns
none recorded
Tools
Tools allowed
Reasoning budget
max
Framework
not recorded
Protocol
Legal Agent Benchmark; full Anthropic evaluation set; mean criterion-pass rate.
Conditions
1,235 tasks after 16 data-defect exclusions; adaptive max effort; public Messages API safeguards/fallback; internal reduced-tool reimplementation; five-run evaluation.