Claude Haiku 5.5 · Medicinal Chemistry ADME(504 drug-discovery questions about measured ADME properties) 58% · 附件证据(ev-h55-card-adme)

Claude Haiku 5.5 的 Medicinal Chemistry ADME · 504 drug-discovery questions about measured ADME properties 官方出处。Life sciences capabilities · 58%。见《Claude Haiku 5.5 System Card》System Card PDF p.138, §8.13.3。证据 ev-h55-card-adme。

原文与定位

Claude Haiku 5.5 · Medicinal Chemistry ADME(504 drug-discovery questions about measured ADME properties),Life sciences capabilities配置,Accuracy 58%(具体成绩公布时间未确认)。见《Claude Haiku 5.5 System Card》System Card PDF p.138, §8.13.3。 查看成绩条件

Claude Haiku 5.5 scored 58.0% on 504 medicinal-chemistry ADME questions.
官方材料
Claude Haiku 5.5 System Card · PDF 第 138 页
定位
附件 · System Card PDF p.138, §8.13.3
材料版本
144 页 PDF/TXT:本地留存副本、current-haiku-5-5-system-card.txt;封面日期 2026-10-07;正文 pp.1-144 已全文核读,全部63个含图/向量图表页实际视觉检查(本轮补看53页,合并此前10页),无未读页或未查看图表页;此为 current 卡片,不作为首发日附件证据。
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 Haiku 5.5 · Medicinal Chemistry ADME(504 drug-discovery questions about measured ADME properties),Life sciences capabilities配置,Accuracy 58%(具体成绩公布时间未确认)。官方出处 Claude Haiku 5.5 System Card(附件,144 页 PDF/TXT:本地留存副本、current-haiku-5-5-system-card.txt;封面日期 2026-10-07;正文 pp.1-144 已全文核读,全部63个含图/向量图表页实际视觉检查(本轮补看53页,合并此前10页),无未读页或未查看图表页;此为 current 卡片,不作为首发日附件证据。)System Card PDF p.138, §8.13.3,https://www.anthropic.com/claude-haiku-5-5-system-card#page=138。Benchlog https://benchlog.lekshome.cc/evidence/claude-haiku-5-5/ev-h55-card-adme/

关联成绩与条件

Claude Haiku 5.5 · Medicinal Chemistry ADME 成绩与配置
指标与成绩配置与评测条件证据关系
Accuracy58%原文 58.0% · 单位 %
公布 未确认
Life sciences capabilities
设置
Adaptive thinking at max effort; biology safeguards disabled; five attempts per problem unless an evaluation subsection states otherwise.
未知项
Per-benchmark sample counts are not stated for all evaluations.
工具
工具条件未记录
推理预算
max
框架
未记录
协议
Medicinal chemistry ADME; 504 questions about measured properties of drug-like molecules; five attempts per problem.
条件
Max effort; biology safeguards disabled; result describes underlying model capability, not deployed safeguards.

直接支持该成绩