GPT Terra · 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. · 历代记录

按发布时间查看 GPT Terra 的 2 次发布在 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. 的成绩、配置与披露。

仅为直接前代间的唯一、有效精确值计算差值,要求单位、指标与明确记录的配置和条件一致。

GPT-5.6 Terra

· 本次发布有 5 条成绩记录,含 5 条精确值(可含历史记录)

GPT-5.6 Terra · 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 error53.6%原文 53.6% · 单位 %
公布
GPT-5.6 Terra low
设置
GPT-5.6 Terra; reasoning effort low, directly labeled in the official chart.
未知项
Sampling and harness settings are not stated for this chart.
工具
工具条件未记录
推理预算
low
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-5.6 Terra; reasoning effort Low; chart cost $0.055.
在排名中查看此成绩
Responses with at least one factual error51.5%原文 51.5% · 单位 %
公布
GPT-5.6 Terra medium
设置
GPT-5.6 Terra; reasoning effort medium, directly labeled in the official chart.
未知项
Sampling and harness settings are not stated for this chart.
工具
工具条件未记录
推理预算
medium
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-5.6 Terra; reasoning effort Medium; chart cost $0.061.
在排名中查看此成绩
Responses with at least one factual error41.1%原文 41.1% · 单位 %
公布
GPT-5.6 Terra high
设置
GPT-5.6 Terra; reasoning effort high, directly labeled in the official chart.
未知项
Sampling and harness settings are not stated for this chart.
工具
工具条件未记录
推理预算
high
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-5.6 Terra; reasoning effort High; chart cost $0.078.
在排名中查看此成绩
Responses with at least one factual error33.9%原文 33.9% · 单位 %
公布
GPT-5.6 Terra xhigh
设置
GPT-5.6 Terra; reasoning effort xhigh, directly labeled in the official chart.
未知项
Sampling and harness settings are not stated for this chart.
工具
工具条件未记录
推理预算
xhigh
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-5.6 Terra; reasoning effort Xhigh; chart cost $0.11.
在排名中查看此成绩
Responses with at least one factual error22.3%原文 22.3% · 单位 %
公布
GPT-5.6 Terra max
设置
GPT-5.6 Terra; reasoning effort max, directly labeled in the official chart.
未知项
Sampling and harness settings are not stated for this chart.
工具
工具条件未记录
推理预算
max
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-5.6 Terra; reasoning effort Max; chart cost $0.25.
在排名中查看此成绩

GPT-6.1 Terra

· 本次发布有 5 条成绩记录,含 5 条精确值(可含历史记录)

与直接前代的记录缺少满足差值计算的同条件依据。

GPT-6.1 Terra · 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 error16.7%原文 16.7% · 单位 %
公布 未确认
GPT-6.1 Terra low
设置
GPT-6.1 Terra; reasoning effort low, directly labeled in the official chart.
未知项
Sampling and harness settings are not stated for this chart.
工具
工具条件未记录
推理预算
low
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-6.1 Terra; reasoning effort Low; chart cost $0.024.
在排名中查看此成绩
Responses with at least one factual error10.1%原文 10.1% · 单位 %
公布 未确认
GPT-6.1 Terra medium
设置
GPT-6.1 Terra; reasoning effort medium, directly labeled in the official chart.
未知项
Sampling and harness settings are not stated for this chart.
工具
工具条件未记录
推理预算
medium
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-6.1 Terra; reasoning effort Medium; chart cost $0.029.
在排名中查看此成绩
Responses with at least one factual error7.7%原文 7.7% · 单位 %
公布 未确认
GPT-6.1 Terra high
设置
GPT-6.1 Terra; reasoning effort high, directly labeled in the official chart.
未知项
Sampling and harness settings are not stated for this chart.
工具
工具条件未记录
推理预算
high
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-6.1 Terra; reasoning effort High; chart cost $0.036.
在排名中查看此成绩
Responses with at least one factual error5%原文 5.0% · 单位 %
公布 未确认
GPT-6.1 Terra xhigh
设置
GPT-6.1 Terra; reasoning effort xhigh, directly labeled in the official chart.
未知项
Sampling and harness settings are not stated for this chart.
工具
工具条件未记录
推理预算
xhigh
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-6.1 Terra; reasoning effort Xhigh; chart cost $0.05.
在排名中查看此成绩
Responses with at least one factual error4.3%原文 4.3% · 单位 %
公布 未确认
GPT-6.1 Terra max
设置
GPT-6.1 Terra at maximum reasoning effort, as stated in the release article for Terminal-Bench 4.0.
未知项
Other sampling and harness settings are not disclosed in the release article.
工具
工具条件未记录
推理预算
maximum
框架
未记录
协议
Factual error rate on difficult prompts (lower is better); y metric Answers with any factual error; x axis Cost (USD). Lower is better; article describes an internal benchmark of error-inducing prompts without identifying the user-flagged dataset.
条件
GPT-6.1 Terra; reasoning effort Max; chart cost $0.12.
在排名中查看此成绩