Gemini 2.5 Pro (GA) · SWE-bench Verified(Verified subset (500 problems)) 67.2% · 附件证据(hist-t3-swe-multi)

Gemini 2.5 Pro (GA) 的 SWE-bench Verified · Verified subset (500 problems) 官方出处。Gemini 2.5 Pro stable API · 67.2%。见《Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities — launch-day 70-page version》Historical PDF p.12, Table 3, SWE-bench Verified / multiple attempts row, Gemini 2.5 Pro column。证据 hist-t3-swe-multi。

原文与定位

Gemini 2.5 Pro (GA) · SWE-bench Verified(Verified subset (500 problems)),Gemini 2.5 Pro stable API配置,Pass rate 67.2%(具体成绩公布时间未确认)。见《Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities — launch-day 70-page version》Historical PDF p.12, Table 3, SWE-bench Verified / multiple attempts row, Gemini 2.5 Pro column。 查看成绩条件

Gemini 2.5 Pro / swe-bench-verified: 67.2%. Multiple agentic traces sampled and reranked using Gemini's own judgement; temperature=1, top-p=0.99, top-k=1024.
官方材料
Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities — launch-day 70-page version · PDF 第 12 页
定位
附件 · Historical PDF p.12, Table 3, SWE-bench Verified / multiple attempts row, Gemini 2.5 Pro column
材料版本
70-page PDF, archived capture from 2025-06-17; cover dated 2025-06-17
版本公布
未确认
披露时间
· 已有首发依据
采集时间
历史依据
已实际下载首发当天官方PDF回放 https://web.archive.org/web/20250617164147id_/https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf;Memento-Datetime Tue, 17 Jun 2025 16:41:47 GMT。70/70页全文、安全章节及附录已核读,成绩表和图表已视觉核查;与现行73页版分开保存。
引用文本

Gemini 2.5 Pro (GA) · SWE-bench Verified(Verified subset (500 problems)),Gemini 2.5 Pro stable API配置,Pass rate 67.2%(具体成绩公布时间未确认)。官方出处 Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities — launch-day 70-page version(附件,70-page PDF, archived capture from 2025-06-17; cover dated 2025-06-17)Historical PDF p.12, Table 3, SWE-bench Verified / multiple attempts row, Gemini 2.5 Pro column,https://web.archive.org/web/20250617164147id_/https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf#page=12。Benchlog https://benchlog.lekshome.cc/evidence/gemini-2-5-pro/hist-t3-swe-multi/

关联成绩与条件

Gemini 2.5 Pro (GA) · SWE-bench Verified 成绩与配置
指标与成绩配置与评测条件证据关系
Pass rate67.2%原文 67.2% · 单位 %
公布 未确认
Gemini 2.5 Pro stable API
设置
AI Studio API model id gemini-2.5-pro; default sampling settings. Technical report states Gemini scores are pass@1 and single-attempt unless specified. Aider Polyglot averages 3 trials; SWE-bench uses the internal agentic harness; Vibe-Eval uses Gemini as judge.
未知项
Thinking budget and any non-default settings for individual evaluations are not stated except where noted in each protocol.
工具
工具条件未记录
推理预算
未记录
框架
未记录
协议
Multiple agentic traces sampled and reranked using Gemini's own judgement; temperature=1, top-p=0.99, top-k=1024.
条件
AI Studio API model gemini-2.5-pro; internal agentic harness.

直接支持该成绩

在排名中查看此成绩