GPT-6.1 Terra · 官方评测披露

GPT-6.1 Terra 当前收录 10 项测试的官方记录。已完成列明材料的核查。

模型发布 · GPT Terra

官方材料与核查范围

Official English article body and all six interactive charts were reviewed; chart data points were cross-checked 143/143 against SVG aria-labels and embedded Vega values. The 36-page system-card PDF/HTML, API model page, and RSS date were also reviewed. Launch-time chart availability is tracked separately in the launch review. 核对时间 。

可靠首发材料中已记录 0 项测试。 Wayback 于 2026-10-26 19:49:19Z 保存了可回放的英文发布页,正文中含 Terminal-Bench 4.0 max 50.3%;交互图表未恢复,且当日未找到系统卡快照,不能将首发材料核查标为完整。最早 18:05:14Z 的 200 快照回放请求被 Wayback 429 拒绝。

  • Introducing GPT-6.1 Terra | OpenAI · 主文 · Official GPT-6.1 Terra article; body and all six interactive charts reviewed; exact chart points cross-checked against official SVG aria-labels and embedded Vega values (143/143). The chart's launch-day inclusion is not verified.
    公布 · 采集 · 已核查 Full article body and all six charts reviewed; exact points for this model were read from the official chart data and aria-label evidence.
  • Addendum to GPT-6 Astra System Card: GPT-6.1 Terra · 附件 · Official HTML addendum; all 36 PDF-matched pages and embedded tables/figures reviewed; current page version date not stated.
    公布 未确认 · 采集 · 已核查 官方 HTML addendum 与 PDF 对照核读;覆盖全部 36 页正文及图表。仅将常规能力 HealthBench 表纳入成绩,安全/对齐/Preparedness/监控实验未混入常规能力。
  • Addendum to GPT-6 Astra System Card: GPT-6.1 Terra (PDF) · 附件 · 36-page official PDF; all extracted pages and rendered figures reviewed. PDF publication/version date not stated.
    公布 未确认 · 采集 · 已核查 36 页 PDF 全文抽取并逐页渲染;所有表格与图表已视觉检查,HealthBench p.9/Table(印刷页 8)逐项核值。
  • GPT-6.1 Terra Model | OpenAI API · 主文 · Current official API model page; confirms public ID and general use description; page revision date unknown.
    公布 未确认 · 采集 · 已核查 确认公开模型 ID、用途与支持 reasoning.effort;API 文档没有 benchmark 成绩。
  • OpenAI News RSS · 主文 · Current RSS; item confirms the official release publication date.
    公布 未确认 · 采集 · 已核查 OpenAI RSS 项目日期为 2026-10-26;用于确认发布日期。

配置

  • GPT-6.1 Terra max · GPT-6.1 Terra at maximum reasoning effort, as stated in the release article for Terminal-Bench 4.0. maximum
    未知项 Other sampling and harness settings are not disclosed in the release article.
  • GPT-6.1 Terra, system-card comparison configuration unspecified · GPT-6.1 Terra values from the system-card HealthBench comparison table; reasoning effort and serving configuration are not stated.
    未知项 Reasoning effort not stated、Sampling and tool settings not stated
  • GPT-6.1 Terra low · GPT-6.1 Terra; reasoning effort low, directly labeled in the official chart. low
    未知项 Sampling and harness settings are not stated for this chart.
  • GPT-6.1 Terra medium · GPT-6.1 Terra; reasoning effort medium, directly labeled in the official chart. medium
    未知项 Sampling and harness settings are not stated for this chart.
  • GPT-6.1 Terra high · GPT-6.1 Terra; reasoning effort high, directly labeled in the official chart. high
    未知项 Sampling and harness settings are not stated for this chart.
  • GPT-6.1 Terra xhigh · GPT-6.1 Terra; reasoning effort xhigh, directly labeled in the official chart. xhigh
    未知项 Sampling and harness settings are not stated for this chart.

直接前代与后续发布

直接前代 GPT-5.6 Terra。 官方关系依据

A significant step up from GPT-5.6 Terra.
6 项披露延续 · 4 项新增 · 23 项本次未见 · 0 项待核查
GPT-5.6 Terra 与 GPT-6.1 Terra 的披露变化
测试已核查状态GPT-5.6 TerraGPT-6.1 Terra
Terminal-Bench · 4.0延续

已记录的评测条件有变化

前代记录本次记录
HealthBench Professional · 525 physician-authored conversations延续

来源位置 main → attachment

已记录的评测条件有变化

前代记录本次记录
HealthBench · 48,000+ rubric items across 5,000 multi-turn conversations新增—本次记录
HealthBench Hard · System-card evaluation; exact dataset version unspecified新增—本次记录
HealthBench Consensus · System-card evaluation; exact dataset version unspecified新增—本次记录
Agents’ Last Exam · V1; 55 subfields; exact task snapshot unspecified延续

已记录的评测条件有变化

前代记录本次记录
MMMU-Pro · Vision format explicitly labeled by the official comparison chart; task release/version unspecified延续

已记录的评测条件有变化

前代记录本次记录
OSWorld · 2.1 offline set; task count and exact task release unspecified in the selected GPT-6.1 Terra article/chart延续

已记录的评测条件有变化

前代记录本次记录
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.延续

已记录的评测条件有变化

前代记录本次记录
Terminal-Bench-Science · 0.1新增—本次记录
Agents’ Last Exam · 55 professional fields; exact task version unspecified本次未见前代记录—
GDPval-AA · v2本次未见前代记录—
Big Finance Bench · Version and task protocol unspecified本次未见前代记录—
Artificial Analysis Intelligence Index · v4.1本次未见前代记录—
Artificial Analysis Coding Agent Index · v1.1本次未见前代记录—
SWE-Bench Pro · Subset/version unspecified in source本次未见前代记录—
DeepSWE · v1.1本次未见前代记录—
Terminal-Bench · 2.1; source metric reported as score percentage本次未见前代记录—
GeneBench Pro · Version unspecified in source本次未见前代记录—
LifeSciBench · Version unspecified in source本次未见前代记录—
OSWorld · 2.0; subset and scoring details unspecified本次未见前代记录—
BrowseComp · 1,266 题;单代理与多代理按协议区分本次未见前代记录—
BenchCAD · Task version and scoring details unspecified本次未见前代记录—
BenchCAD · Task version unspecified; Python tool protocol本次未见前代记录—
MMMU-Pro · Standard and Vision formats本次未见前代记录—
MMMU-Pro · Tool-enabled protocol; exact version unspecified (gpt56-mmmu-pro-with-tools)本次未见前代记录—
GDP.pdf · 100 professional PDFs and prompts本次未见前代记录—
GPQA Diamond · 198-question Diamond subset本次未见前代记录—
FrontierMath · Tier 1–3 v2本次未见前代记录—
FrontierMath · Tier 4 v2本次未见前代记录—
Toolathlon · Version, subset, and metric unspecified in the 2026-07-09 article本次未见前代记录—
ARC-AGI-3 · Version/subset unspecified in source本次未见前代记录—
AutomationBench · Zapier 私有 early-access set;版本未注明本次未见前代记录—

测试记录 · 10