GPT-5.6 Luna · BrowseComp · 1,266 题;单代理与多代理按协议区分 · 成绩与配置

GPT-5.6 Luna 的 BrowseComp · 1,266 题;单代理与多代理按协议区分 记录。本次发布有 6 条成绩记录,含 6 条精确值(可含历史记录),附 6 条来源定位。

成绩与评测条件

GPT-5.6 Luna · BrowseComp 成绩与配置
指标与成绩配置与评测条件出处
Accuracy83.3%原文 83.3% · 单位 %
公布 未确认
GPT-5.6 Luna · 档位未注明
设置
主文表格未逐行标明本型号推理档位;不套用 Sol 的 effort,也不以价格推断档位。
未知项
推理档位、部分评测的完整 harness 与采样设置
工具
工具条件未记录
推理预算
未记录
框架
未记录
协议
未说明
条件
GPT-5.6 Luna;主文同表主体列原值,未逐行标明 effort。行名与工具条件:BrowseComp。
在排名中查看此成绩
Accuracy45.73%原文 45.73% · 单位 %
公布 未确认
GPT-5.6 Luna (low)
设置
GPT-5.6 Luna; reasoning effort Low as directly labeled on the official chart.
未知项
Per-benchmark sampling and chart harness settings are not stated.
工具
工具条件未记录
推理预算
Low
框架
未记录
协议
BrowseComp; chart score; task harness details are not stated in this figure.
条件
GPT-5.6 Luna; reasoning effort Low directly labeled by the chart. Chart x-axis API cost: $0.08. Other run settings are not stated.
在排名中查看此成绩
Accuracy58.85%原文 58.85% · 单位 %
公布 未确认
GPT-5.6 Luna (medium)
设置
GPT-5.6 Luna; reasoning effort Medium as directly labeled on the official chart.
未知项
Per-benchmark sampling and chart harness settings are not stated.
工具
工具条件未记录
推理预算
Medium
框架
未记录
协议
BrowseComp; chart score; task harness details are not stated in this figure.
条件
GPT-5.6 Luna; reasoning effort Medium directly labeled by the chart. Chart x-axis API cost: $0.23. Other run settings are not stated.
在排名中查看此成绩
Accuracy80.02%原文 80.02% · 单位 %
公布 未确认
GPT-5.6 Luna (high)
设置
GPT-5.6 Luna; reasoning effort High as directly labeled on the official chart.
未知项
Per-benchmark sampling and chart harness settings are not stated.
工具
工具条件未记录
推理预算
High
框架
未记录
协议
BrowseComp; chart score; task harness details are not stated in this figure.
条件
GPT-5.6 Luna; reasoning effort High directly labeled by the chart. Chart x-axis API cost: $1.03. Other run settings are not stated.
在排名中查看此成绩
Accuracy84.04%原文 84.04% · 单位 %
公布 未确认
GPT-5.6 Luna (xhigh)
设置
GPT-5.6 Luna; reasoning effort Xhigh as directly labeled on the official chart.
未知项
Per-benchmark sampling and chart harness settings are not stated.
工具
工具条件未记录
推理预算
Xhigh
框架
未记录
协议
BrowseComp; chart score; task harness details are not stated in this figure.
条件
GPT-5.6 Luna; reasoning effort Xhigh directly labeled by the chart. Chart x-axis API cost: $1.33. Other run settings are not stated.
在排名中查看此成绩
Accuracy83.25%原文 83.25% · 单位 %
公布 未确认
GPT-5.6 Luna (max)
设置
GPT-5.6 Luna; reasoning effort Max as directly labeled on the official chart.
未知项
Per-benchmark sampling and chart harness settings are not stated.
工具
工具条件未记录
推理预算
Max
框架
未记录
协议
BrowseComp; chart score; task harness details are not stated in this figure.
条件
GPT-5.6 Luna; reasoning effort Max directly labeled by the chart. Chart x-axis API cost: $1.42. Other run settings 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 BrowseComp、Multi-agent BrowseComp

  • Accuracy · % · 越高越好