GPT-6.2 Sol · User-flagged factuality(Refreshed 1,000 de-identified conversations: 229 retained cases and 771 new cases; separate general factuality and reported-issue graders; not comparable with earlier system-card versions) · 附件证据(ev-gpt62-sol-user-flagged-factuality-1000-figure25)

GPT-6.2 Sol 的 User-flagged factuality · Refreshed 1,000 de-identified conversations: 229 retained cases and 771 new cases; separate general factuality and reported-issue graders; not comparable with earlier system-card versions 官方出处。见《GPT-6.2 Astra and Sol System Card》HTML §3.3.1 Performance in Cases Flagged by Users / Figure 25; PDF physical pages 72–73 / printed pages 71–72 / §3.3.0.1. 图中没有精确点值标签,HTML仅嵌入静态PNG,不按坐标估值。。证据 ev-gpt62-sol-user-flagged-factuality-1000-figure25。

原文与定位

Performance in Cases Flagged by Users is an internal OpenAI evaluation that tests whether a model avoids factual errors reported by users of earlier models. The current version contains 1,000 de-identified conversations and covers both incorrect claims about the world and mistakes in interpreting information supplied by the user, including text and images. We refreshed this evaluation after earlier versions became less informative for recent models. The current version retains 229 cases from the previous version and adds 771 cases drawn from newer conversations. We reran the comparison models on this version, and its results should not be compared directly with scores from earlier versions as reported in previous system cards. GPT-6.2 Astra and GPT-6.2 Sol achieved lower error rates than GPT-6 Astra at comparable simulated response times, with error rates tending to decline as they spent more time reasoning. Figure 25. Response-level hallucination rates and rates of reproducing the user-reported error on 1,000 selected conversations, plotted against mean simulated response time. Each point represents a tested reasoning setting. The two panels use separate automated graders; lower rates indicate better performance.
官方材料
GPT-6.2 Astra and Sol System Card · 定位到章节
定位
附件 · HTML §3.3.1 Performance in Cases Flagged by Users / Figure 25; PDF physical pages 72–73 / printed pages 71–72 / §3.3.0.1. 图中没有精确点值标签,HTML仅嵌入静态PNG,不按坐标估值。
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 · User-flagged factuality(Refreshed 1,000 de-identified conversations: 229 retained cases and 771 new cases; separate general factuality and reported-issue graders; not comparable with earlier system-card versions),披露记录。官方出处 GPT-6.2 Astra and Sol System Card(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。)HTML §3.3.1 Performance in Cases Flagged by Users / Figure 25; PDF physical pages 72–73 / printed pages 71–72 / §3.3.0.1. 图中没有精确点值标签,HTML仅嵌入静态PNG,不按坐标估值。,https://deploymentsafety.openai.com/gpt-6-2-astra-sol#performance-in-cases-flagged-by-users。Benchlog https://benchlog.lekshome.cc/evidence/gpt-6-2-sol/ev-gpt62-sol-user-flagged-factuality-1000-figure25/

关联成绩与条件

已收录官方披露,暂无成绩记录。