GPT-5.1 Codex-Max · MLE-Bench(30 competition subset selected from tasks with <50GB data and <10h reference duration) 17% · 附件证据(ev-gpt51-codexmax-mlebench-card)

GPT-5.1 Codex-Max 的 MLE-Bench · 30 competition subset selected from tasks with <50GB data and <10h reference duration 官方出处。System Card reported effort · 17%。见《GPT-5.1-Codex-Max System Card PDF | OpenAI》Official System Card PDF p.23, Figure 12; numeric bar label visually inspected。证据 ev-gpt51-codexmax-mlebench-card。

原文与定位

GPT-5.1 Codex-Max · MLE-Bench(30 competition subset selected from tasks with <50GB data and <10h reference duration),System Card reported effort配置,Bronze pass@1 17%(具体成绩公布时间未确认)。见《GPT-5.1-Codex-Max System Card PDF | OpenAI》Official System Card PDF p.23, Figure 12; numeric bar label visually inspected。 查看成绩条件

Figure 12: gpt-5.1-codex-max: 17%. MLE-bench-30 selected subset; <50GB data and <10h reference-task filter; bronze pass@1; agent has a GPU and Kaggle data. Generic environment description allows 24h and some 100h experiments; exact per-model duration unspecified.
官方材料
GPT-5.1-Codex-Max System Card PDF | OpenAI · PDF 第 23 页
定位
附件 · Official System Card PDF p.23, Figure 12; numeric bar label visually inspected
材料版本
Directly linked PDF downloaded from the official CDN; full text and all figure pages are being reviewed. Current HTML/PDF fully read; PDF has 27 pages. November 18 is the displayed publication/cover date, not proof of launch bytes; exact current revision date unknown.
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-5.1 Codex-Max · MLE-Bench(30 competition subset selected from tasks with <50GB data and <10h reference duration),System Card reported effort配置,Bronze pass@1 17%(具体成绩公布时间未确认)。官方出处 GPT-5.1-Codex-Max System Card PDF | OpenAI(附件,Directly linked PDF downloaded from the official CDN; full text and all figure pages are being reviewed. Current HTML/PDF fully read; PDF has 27 pages. November 18 is the displayed publication/cover date, not proof of launch bytes; exact current revision date unknown.)Official System Card PDF p.23, Figure 12; numeric bar label visually inspected,https://cdn.openai.com/pdf/2a7d98b1-57e5-4147-8d0e-683894d782ae/5p1_codex_max_card_03.pdf#page=23。Benchlog https://benchlog.lekshome.cc/evidence/gpt-5-1-codex-max/ev-gpt51-codexmax-mlebench-card/

关联成绩与条件

GPT-5.1 Codex-Max · MLE-Bench 成绩与配置
指标与成绩配置与评测条件证据关系
Bronze pass@117%原文 17% · 单位 %
公布 未确认
System Card reported effort
设置
GPT-5.1-Codex-Max; benchmark-specific single-agent setup; effort not stated per figure.
未知项
SWE-Lancer与MLE-Bench图没有单独注明推理档位、采样参数或运行数;MLE具体运行时长未逐模型给出。
工具
工具条件未记录
推理预算
未记录
框架
未记录
协议
MLE-bench-30 selected subset; <50GB data and <10h reference-task filter; bronze pass@1; agent has a GPU and Kaggle data. Generic environment description allows 24h and some 100h experiments; exact per-model duration unspecified.
条件
原图明示百分比;未从柱高估数。GPT-5.1对照值属于后续来源。卡片当前修订时点未确认,publishedAt保持null;不混入内部PR/Proof Q&A或风险子任务。

直接支持该成绩

在排名中查看此成绩