Claude Sonnet 5 · Legal Agent Benchmark(Public full set) · 附件证据(ev5-lab-public)

Claude Sonnet 5 的 Legal Agent Benchmark · Public full set 官方出处。见《Claude Sonnet 5 System Card》§8.11.3, PDF p.133。证据 ev5-lab-public。

原文与定位

Claude Sonnet 5 · Legal Agent Benchmark(Public full set),Standard configuration: adaptive thinking, max effort配置,All-pass rate 8.92%(具体成绩公布时间未确认)。见《Claude Sonnet 5 System Card》§8.11.3, PDF p.133。 查看成绩条件

Claude Sonnet 5 · Legal Agent Benchmark(Public full set),Standard configuration: adaptive thinking, max effort配置,Mean criterion-pass rate 88.26%(具体成绩公布时间未确认)。见《Claude Sonnet 5 System Card》§8.11.3, PDF p.133。 查看成绩条件

We tested Claude Sonnet 5 against 1,235 problems (16 of the 1,251 problems were excluded due to data defects; exclusions were identified before testing) and achieved 8.92% (± 0.36, n=5) all-pass rate and 88.26% mean criterion-pass rate (adaptive thinking, max effort).
官方材料
Claude Sonnet 5 System Card · PDF 第 133 页
定位
附件 · §8.11.3, PDF p.133
材料版本
PDF with changelog dated 2026-07-10; 146 pages; collected 2026-11-10
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 Sonnet 5 · Legal Agent Benchmark(Public full set),Standard configuration: adaptive thinking, max effort配置,All-pass rate 8.92%(具体成绩公布时间未确认);Claude Sonnet 5 · Legal Agent Benchmark(Public full set),Standard configuration: adaptive thinking, max effort配置,Mean criterion-pass rate 88.26%(具体成绩公布时间未确认)。官方出处 Claude Sonnet 5 System Card(附件,PDF with changelog dated 2026-07-10; 146 pages; collected 2026-11-10)§8.11.3, PDF p.133,https://www.anthropic.com/claude-sonnet-5-system-card#page=133。Benchlog https://benchlog.lekshome.cc/evidence/claude-sonnet-5/ev5-lab-public/

关联成绩与条件

Claude Sonnet 5 · Legal Agent Benchmark 成绩与配置
指标与成绩配置与评测条件证据关系
All-pass rate8.92%原文 8.92% (±0.36, n=5) · 单位 %
公布 未确认
Standard configuration: adaptive thinking, max effort
设置
Adaptive thinking at max effort; default temperature and top_p; five-trial averages unless otherwise stated.
未知项
Context-window size is evaluation-dependent; standard configuration is 1M tokens.
工具
允许工具
推理预算
max
框架
未记录
协议
Legal Agent Benchmark public tasks, 1,235 problems after excluding 16 data defects; five runs.
条件
Adaptive thinking, max effort; pass requires all rubric criteria. Internal harness uses Sonnet 4.6 as judge and only bash and Python tools, versus more tools in the public harness.

直接支持该成绩

在排名中查看此成绩
Mean criterion-pass rate88.26%原文 88.26% · 单位 %
公布 未确认
Standard configuration: adaptive thinking, max effort
设置
Adaptive thinking at max effort; default temperature and top_p; five-trial averages unless otherwise stated.
未知项
Context-window size is evaluation-dependent; standard configuration is 1M tokens.
工具
工具条件未记录
推理预算
max
框架
未记录
协议
Legal Agent Benchmark public tasks; mean criterion-pass rate.
条件
Adaptive thinking, max effort; 1,235 problems, five runs.

直接支持该成绩

在排名中查看此成绩