GPT-6 Astra · HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations · 成绩与配置

GPT-6 Astra 的 HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations 记录。本次发布有 3 条成绩记录,含 3 条精确值(可含历史记录),附 2 条来源定位。

成绩与评测条件

GPT-6 Astra · HealthBench 成绩与配置
指标与成绩配置与评测条件出处
长度校正分(0–100)58.3 score原文 58.3 · 单位 score
公布

System Card changelog dated 2026-09-22: HealthBench values for GPT-6 Astra were revised to correct a misconfiguration in previous evaluations.

GPT-6 Astra HealthBench corrected system-card values
设置
System-card values revised on 2026-09-22 to correct misconfiguration in previous evaluations.
未知项
Previous exact value mapping not disclosed in the changelog
工具
工具条件未记录
推理预算
未记录
框架
未记录
协议
OpenAI 系统卡 HealthBench 计分;2,000 字符不调整,每多 500 字符减 2.99 分,较短回复对应加分;grader 未说明。
条件
Table 6 current reported row; exact reasoning effort and sampling are not stated. The card changelog identifies HealthBench values as revised for a prior misconfiguration; no unique prior-value mapping is given. 系统卡明确采用 0–100 量表;保留原始分数字符串,不乘除数值。
在排名中查看此成绩
未校正分(0–100)56.9 score原文 56.9 · 单位 score
公布

System Card changelog dated 2026-09-22: HealthBench values for GPT-6 Astra were revised to correct a misconfiguration in previous evaluations.

GPT-6 Astra HealthBench corrected system-card values
设置
System-card values revised on 2026-09-22 to correct misconfiguration in previous evaluations.
未知项
Previous exact value mapping not disclosed in the changelog
工具
工具条件未记录
推理预算
未记录
框架
未记录
协议
OpenAI 系统卡 HealthBench 计分;2,000 字符不调整,每多 500 字符减 2.99 分,较短回复对应加分;grader 未说明。
条件
Exact reasoning effort and sampling are not stated. This is the unadjusted score paired with the length-adjusted value in Table 6. 系统卡明确采用 0–100 量表;保留原始分数字符串,不乘除数值。
在排名中查看此成绩
Mean response length1760 characters原文 1760 · 单位 characters
公布

System Card changelog dated 2026-09-22: HealthBench values for GPT-6 Astra were revised to correct a misconfiguration in previous evaluations.

GPT-6 Astra HealthBench corrected system-card values
设置
System-card values revised on 2026-09-22 to correct misconfiguration in previous evaluations.
未知项
Previous exact value mapping not disclosed in the changelog
工具
工具条件未记录
推理预算
未记录
框架
未记录
协议
OpenAI 系统卡 HealthBench 计分;2,000 字符不调整,每多 500 字符减 2.99 分,较短回复对应加分;grader 未说明。
条件
Length statistic paired with the length-adjusted and unadjusted scores; exact reasoning effort and sampling are not stated.
在排名中查看此成绩

出处与披露

测试版本与指标

别名 HealthBench

  • Length-adjusted score · % · 越高越好
  • Raw score · % · 越高越好
  • Mean response length · characters · 排序方向未确认
  • 长度校正分(0–100) · score · 越高越好
  • 未校正分(0–100) · score · 越高越好
  • 相对 GPT-5.4 的长度校正分差 · points · 排序方向未确认
  • Unadjusted score · % · 越高越好