GPT-6 Astra · Factual error rate on difficult prompts · Internal error-inducing prompts in the GPT-6.1 Terra comparison article; exact dataset version, sample size and user-flagged provenance unspecified. · 成绩与配置

GPT-6 Astra 的 Factual error rate on difficult prompts · Internal error-inducing prompts in the GPT-6.1 Terra comparison article; exact dataset version, sample size and user-flagged provenance unspecified. 记录。本次发布有 5 条成绩记录,含 5 条精确值(可含历史记录),附 5 条来源定位。

成绩与评测条件

GPT-6 Astra · Factual error rate on difficult prompts 成绩与配置
指标与成绩配置与评测条件出处
Responses with at least one factual error13.3%原文 13.3% · 单位 %
公布 未确认
GPT-6 Astra low
设置
GPT-6 Astra; reasoning effort low as printed in the interactive chart.
未知项
未另记
工具
工具条件未记录
推理预算
low
框架
未记录
协议
Internal error-inducing difficult prompts; responses with any factual error; lower is better. Exact dataset, sample size and user-flagged provenance not stated in this source.
条件
Later GPT-6.1 Terra comparison article; GPT-6 Astra effort Low. Chart Cost: USD 0.26 per task; exact cost-accounting and sampling settings unspecified. All 143 chart points were cross-checked against official embedded values and rendered SVG labels. This result is attributed to Astra, not to the article's Terra subject; no launch qualification is inferred.
在排名中查看此成绩
Responses with at least one factual error9.7%原文 9.7% · 单位 %
公布 未确认
GPT-6 Astra medium
设置
GPT-6 Astra; reasoning effort medium as printed in the interactive chart.
未知项
未另记
工具
工具条件未记录
推理预算
medium
框架
未记录
协议
Internal error-inducing difficult prompts; responses with any factual error; lower is better. Exact dataset, sample size and user-flagged provenance not stated in this source.
条件
Later GPT-6.1 Terra comparison article; GPT-6 Astra effort Medium. Chart Cost: USD 0.29 per task; exact cost-accounting and sampling settings unspecified. All 143 chart points were cross-checked against official embedded values and rendered SVG labels. This result is attributed to Astra, not to the article's Terra subject; no launch qualification is inferred.
在排名中查看此成绩
Responses with at least one factual error5.9%原文 5.9% · 单位 %
公布 未确认
GPT-6 Astra high
设置
GPT-6 Astra; reasoning effort high as printed in the interactive chart.
未知项
未另记
工具
工具条件未记录
推理预算
high
框架
未记录
协议
Internal error-inducing difficult prompts; responses with any factual error; lower is better. Exact dataset, sample size and user-flagged provenance not stated in this source.
条件
Later GPT-6.1 Terra comparison article; GPT-6 Astra effort High. Chart Cost: USD 0.43 per task; exact cost-accounting and sampling settings unspecified. All 143 chart points were cross-checked against official embedded values and rendered SVG labels. This result is attributed to Astra, not to the article's Terra subject; no launch qualification is inferred.
在排名中查看此成绩
Responses with at least one factual error4.6%原文 4.6% · 单位 %
公布 未确认
GPT-6 Astra xhigh
设置
GPT-6 Astra; reasoning effort xhigh as printed in the interactive chart.
未知项
未另记
工具
工具条件未记录
推理预算
xhigh
框架
未记录
协议
Internal error-inducing difficult prompts; responses with any factual error; lower is better. Exact dataset, sample size and user-flagged provenance not stated in this source.
条件
Later GPT-6.1 Terra comparison article; GPT-6 Astra effort Xhigh. Chart Cost: USD 0.55 per task; exact cost-accounting and sampling settings unspecified. All 143 chart points were cross-checked against official embedded values and rendered SVG labels. This result is attributed to Astra, not to the article's Terra subject; no launch qualification is inferred.
在排名中查看此成绩
Responses with at least one factual error4.2%原文 4.2% · 单位 %
公布 未确认
GPT-6 Astra max
设置
GPT-6 Astra; reasoning effort max as printed in the interactive chart.
未知项
未另记
工具
工具条件未记录
推理预算
max
框架
未记录
协议
Internal error-inducing difficult prompts; responses with any factual error; lower is better. Exact dataset, sample size and user-flagged provenance not stated in this source.
条件
Later GPT-6.1 Terra comparison article; GPT-6 Astra effort Max. Chart Cost: USD 0.66 per task; exact cost-accounting and sampling settings unspecified. All 143 chart points were cross-checked against official embedded values and rendered SVG labels. This result is attributed to Astra, not to the article's Terra subject; no launch qualification is inferred.
在排名中查看此成绩

出处与披露

测试版本与指标

别名 Factual error rate on difficult prompts (lower is better)

  • Responses with at least one factual error · % · 越低越好