Claude Fable 5.2 · Harvey LAB-AA(Artificial Analysis v1.1; 120-task held-out set; hallucination-gated and ungated all-pass rates) · 附件证据(ev-f52-legal-lab-aa)

Claude Fable 5.2 的 Harvey LAB-AA · Artificial Analysis v1.1; 120-task held-out set; hallucination-gated and ungated all-pass rates 官方出处。见《Claude Fable 5.2 and Claude Mythos 5.2 System Card》PDF p.238, §8.14.2, Harvey LAB-AA v1.1。证据 ev-f52-legal-lab-aa。

原文与定位

Claude Fable 5.2 · Harvey LAB-AA(Artificial Analysis v1.1; 120-task held-out set; hallucination-gated and ungated all-pass rates),Adaptive Max配置,Hallucination-gated all-pass rate 7.2%(具体成绩公布时间未确认)。见《Claude Fable 5.2 and Claude Mythos 5.2 System Card》PDF p.238, §8.14.2, Harvey LAB-AA v1.1。 查看成绩条件

Claude Fable 5.2 · Harvey LAB-AA(Artificial Analysis v1.1; 120-task held-out set; hallucination-gated and ungated all-pass rates),Adaptive Max配置,All-pass rate before hallucination gate 10.8%(具体成绩公布时间未确认)。见《Claude Fable 5.2 and Claude Mythos 5.2 System Card》PDF p.238, §8.14.2, Harvey LAB-AA v1.1。 查看成绩条件

On the 120-task held-out set at max effort, Fable 5.2 scored 7.2% hallucination-gated all-pass, versus 10.8% all-pass before hallucination screening. Artificial Analysis used three judges; version 1.1 scores are not comparable to prior versions.
官方材料
Claude Fable 5.2 and Claude Mythos 5.2 System Card · PDF 第 238 页
定位
附件 · PDF p.238, §8.14.2, Harvey LAB-AA v1.1
材料版本
Official System Card; 264-page PDF, cover dated November 9, 2026; current PDF/TXT saved as 本地留存副本 and .txt; PDF pp.1-264 text read page-by-page; all 80 image/vector-figure pages in pp.1-206 were visually checked in 21 sheets, and capability figures in pp.207-254 were visually checked; no unread pages or figures remain.
版本公布
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 Fable 5.2 · Harvey LAB-AA(Artificial Analysis v1.1; 120-task held-out set; hallucination-gated and ungated all-pass rates),Adaptive Max配置,Hallucination-gated all-pass rate 7.2%(具体成绩公布时间未确认);Claude Fable 5.2 · Harvey LAB-AA(Artificial Analysis v1.1; 120-task held-out set; hallucination-gated and ungated all-pass rates),Adaptive Max配置,All-pass rate before hallucination gate 10.8%(具体成绩公布时间未确认)。官方出处 Claude Fable 5.2 and Claude Mythos 5.2 System Card(附件,Official System Card; 264-page PDF, cover dated November 9, 2026; current PDF/TXT saved as 本地留存副本 and .txt; PDF pp.1-264 text read page-by-page; all 80 image/vector-figure pages in pp.1-206 were visually checked in 21 sheets, and capability figures in pp.207-254 were visually checked; no unread pages or figures remain.)PDF p.238, §8.14.2, Harvey LAB-AA v1.1,https://www.anthropic.com/document/claude-fable-5-2-mythos-5-2-system-card#page=238。Benchlog https://benchlog.lekshome.cc/evidence/claude-fable-and-mythos-5-2/ev-f52-legal-lab-aa/

关联成绩与条件

Claude Fable 5.2 · Harvey LAB-AA 成绩与配置
指标与成绩配置与评测条件证据关系
Hallucination-gated all-pass rate7.2%原文 7.2% · 单位 %
公布 未确认
Adaptive Max
设置
Claude Fable 5.2;adaptive thinking always on;Max effort;默认 temperature/top_p;Card 通用表说明通常五次运行平均,个别测试另行注明。
未知项
System Card pp.1-264 文字已完整核读;各benchmark章节未明确披露的运行参数仍未知。
工具
工具条件未记录
推理预算
max
框架
未记录
协议
Harvey LAB-AA v1.1 held-out 120-task set; Artificial Analysis evaluation; max effort; average across three judge models.
条件
Task passes only if every criterion is met and no material hallucination appears in the deliverable.

直接支持该成绩

在排名中查看此成绩
All-pass rate before hallucination gate10.8%原文 10.8% · 单位 %
公布 未确认
Adaptive Max
设置
Claude Fable 5.2;adaptive thinking always on;Max effort;默认 temperature/top_p;Card 通用表说明通常五次运行平均,个别测试另行注明。
未知项
System Card pp.1-264 文字已完整核读;各benchmark章节未明确披露的运行参数仍未知。
工具
工具条件未记录
推理预算
max
框架
未记录
协议
Harvey LAB-AA v1.1 held-out 120-task set; Artificial Analysis evaluation; max effort; average across three judge models.
条件
All-pass before the hallucination gate; version 1.1 results are not comparable with previous versions.

直接支持该成绩

在排名中查看此成绩