GPT-5 · SWE-Lancer(IC Diamond set) 55% · 附件证据(ev-gpt5-swelancer-ic-diamond-pass1)

GPT-5 的 SWE-Lancer · IC Diamond set 官方出处。Preparedness · non-public verbosity · 55%。见《GPT-5 System Card | OpenAI》PDF pp. 38-39, Section 5.1.2.7 and Figure 24。证据 ev-gpt5-swelancer-ic-diamond-pass1。

原文与定位

GPT-5 · SWE-Lancer(IC Diamond set),Preparedness · non-public verbosity配置,Score 55%(成绩公布 2025-08-07)。见《GPT-5 System Card | OpenAI》PDF pp. 38-39, Section 5.1.2.7 and Figure 24。 查看成绩条件

Figure 24, SWE-Lancer IC SWE (preceding section identifies the IC SWE Diamond set), no browsing: gpt-5-thinking 55%.
官方材料
GPT-5 System Card | OpenAI · PDF 第 39 页
定位
附件 · PDF pp. 38-39, Section 5.1.2.7 and Figure 24
材料版本
Actual archived August 7, 2025 launch PDF, 59 pages; text pp.1-59 and all figure/table pages fully reviewed; independent of current 60-page revision. Existing evidence IDs and launch locators retained.
版本公布
披露时间
· 已有首发依据
采集时间
历史依据
Actual launch PDF replay 20250807170752, HTTP 200 application/pdf, digest ORQL27FZQUBZIEDWSFAKB7HOHFDHRPMS; saved 4,842,287-byte 59-page PDF at 2026-11-16-review-gpt-early/本地留存副本; text pp.1-59 and all figure/table pages read. This is not the unrelated one-page placeholder.
引用文本

GPT-5 · SWE-Lancer(IC Diamond set),Preparedness · non-public verbosity配置,Score 55%(成绩公布 2025-08-07)。官方出处 GPT-5 System Card | OpenAI(附件,Actual archived August 7, 2025 launch PDF, 59 pages; text pp.1-59 and all figure/table pages fully reviewed; independent of current 60-page revision. Existing evidence IDs and launch locators retained.)PDF pp. 38-39, Section 5.1.2.7 and Figure 24,https://web.archive.org/web/20250807170752id_/https://cdn.openai.com/pdf/8124a3ce-ab78-4f06-96eb-49ea29ffb52f/gpt5-system-card-aug7.pdf#page=39。Benchlog https://benchlog.lekshome.cc/evidence/gpt-5/ev-gpt5-swelancer-ic-diamond-pass1/

关联成绩与条件

GPT-5 · SWE-Lancer 成绩与配置
指标与成绩配置与评测条件证据关系
Score55%原文 55% · 单位 %
公布
Preparedness · non-public verbosity
设置
High reasoning effort; maximum trained-in verbosity higher than public API high; one attempt per problem; no browsing. GPT-5 System Card current p.37 states that all Preparedness evaluations use this non-public verbosity.
未知项
该verbosity未通过当时公共API提供;不映射为普通High。
工具
工具条件未记录
推理预算
High
框架
未记录
协议
SWE-Lancer IC SWE Diamond, pass@1; no browsing.
条件
High reasoning effort; one attempt per problem; launch-candidate figure label. Card explicitly states all Preparedness evaluations use maximum trained-in verbosity above the API high setting; retained raw, excluded from ordinary presentation.

直接支持该成绩