GPT-6.2 Sol · MentalHealthBench(Overall; 1,215 synthetic conversations; four independent responses per task; clinician-authored rubrics) 50.9% · 附件证据(ev-gpt62-sol-mentalhealthbench-overall-table19)

GPT-6.2 Sol 的 MentalHealthBench · Overall; 1,215 synthetic conversations; four independent responses per task; clinician-authored rubrics 官方出处。GPT-6.2 Sol max · 50.9%。见《GPT-6.2 Astra and Sol System Card (PDF)》PDF physical pages 75–76 / printed pages 74–75 / Figure 26 and Table 19 / overall row / GPT-6.2 Sol。证据 ev-gpt62-sol-mentalhealthbench-overall-table19。

原文与定位

GPT-6.2 Sol · MentalHealthBench(Overall; 1,215 synthetic conversations; four independent responses per task; clinician-authored rubrics),GPT-6.2 Sol max配置,Mean task-clipped score 50.9%(具体成绩公布时间未确认)。见《GPT-6.2 Astra and Sol System Card (PDF)》PDF physical pages 75–76 / printed pages 74–75 / Figure 26 and Table 19 / overall row / GPT-6.2 Sol。 查看成绩条件

At maximum reasoning effort, GPT-6.2 Astra and GPT-6.2 Sol scored 53.3 percent and 50.9 percent on MentalHealthBench, versus 58.7 percent for GPT-6 Astra and 57.9 percent for GPT-6.1 Sol. Table 19: MentalHealthBench scores at maximum reasoning effort. We average four responses per conversation across 1,215 conversations. The values after ± show one standard error (in percentage points) across conversations. Higher is better.
官方材料
GPT-6.2 Astra and Sol System Card (PDF) · PDF 第 76 页
定位
附件 · PDF physical pages 75–76 / printed pages 74–75 / Figure 26 and Table 19 / overall row / GPT-6.2 Sol
材料版本
官方HTML及119页PDF,封面和change log日期2026-11-17;当前页面后续修订时间未知。119页全文(含脚注和参考文献)已逐页核读,21张表和56张图均已视觉核对;HTML全部119个section与PDF正文及表图对应已核对。安全、Preparedness、对齐与monitorability专项已按用途登记,不自动并入常规能力展示。
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

GPT-6.2 Sol · MentalHealthBench(Overall; 1,215 synthetic conversations; four independent responses per task; clinician-authored rubrics),GPT-6.2 Sol max配置,Mean task-clipped score 50.9%(具体成绩公布时间未确认)。官方出处 GPT-6.2 Astra and Sol System Card (PDF)(附件,官方HTML及119页PDF,封面和change log日期2026-11-17;当前页面后续修订时间未知。119页全文(含脚注和参考文献)已逐页核读,21张表和56张图均已视觉核对;HTML全部119个section与PDF正文及表图对应已核对。安全、Preparedness、对齐与monitorability专项已按用途登记,不自动并入常规能力展示。)PDF physical pages 75–76 / printed pages 74–75 / Figure 26 and Table 19 / overall row / GPT-6.2 Sol,https://deploymentsafety.openai.com/gpt-6-2-astra-sol/gpt-6-2-astra-sol.pdf#page=76。Benchlog https://benchlog.lekshome.cc/evidence/gpt-6-2-sol/ev-gpt62-sol-mentalhealthbench-overall-table19/

关联成绩与条件

GPT-6.2 Sol · MentalHealthBench 成绩与配置
指标与成绩配置与评测条件证据关系
Mean task-clipped score50.9%原文 50.9% ± 1.0 · 单位 %
公布 未确认
GPT-6.2 Sol max
设置
GPT-6.2 Sol; reasoning effort max, explicitly labeled in the official chart data.
未知项
文章未注明逐项采样设置、完整执行框架和工具配置;按图表所示条件记录。
工具
工具条件未记录
推理预算
max
框架
未记录
协议
Four independent responses per conversation; rubric score clipped to 0–100%; within-task mean then mean across 1,215 tasks.
条件
Maximum reasoning effort explicitly stated. 1,215 synthetic conversations with clinician-authored rubrics; overall score, standard error ±1.0 percentage point across task means. 不将non-acute/high-acuity/emergent子集或动态对抗式政策合规评测混为总体成绩。其他采样参数、工具和完整harness未注明。

直接支持该成绩

在排名中查看此成绩