Gemini 2.5 Pro (GA) · ActivityNet-QA(Test subset) 66.7 · 附件证据(t6-activitynet)

Gemini 2.5 Pro (GA) 的 ActivityNet-QA · Test subset 官方出处。Gemini 2.5 Pro stable API · 66.7。见《Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities》PDF p.15, Table 6, ActivityNet-QA row, Gemini 2.5 Pro column。证据 t6-activitynet。

原文与定位

Gemini 2.5 Pro (GA) · ActivityNet-QA(Test subset),Gemini 2.5 Pro stable API配置,Accuracy 66.7(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》PDF p.15, Table 6, ActivityNet-QA row, Gemini 2.5 Pro column。 查看成绩条件

ActivityNet-QA 56.2 57.3 55.3 56.4 65.1 66.7 60.4
官方材料
Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities · PDF 第 15 页
定位
附件 · PDF p.15, Table 6, ActivityNet-QA row, Gemini 2.5 Pro column
材料版本
73-page PDF; embedded results include Gemini 2.5 Pro GA up to version 06-17; PDF metadata creation 2025-10-16 is not treated as publication date
版本公布
未确认
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

Gemini 2.5 Pro (GA) · ActivityNet-QA(Test subset),Gemini 2.5 Pro stable API配置,Accuracy 66.7(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(附件,73-page PDF; embedded results include Gemini 2.5 Pro GA up to version 06-17; PDF metadata creation 2025-10-16 is not treated as publication date)PDF p.15, Table 6, ActivityNet-QA row, Gemini 2.5 Pro column,https://storage.googleapis.com/deepmind-media/gemini/gemini_v2_5_report.pdf#page=15。Benchlog https://benchlog.lekshome.cc/evidence/gemini-2-5-pro/t6-activitynet/

关联成绩与条件

Gemini 2.5 Pro (GA) · ActivityNet-QA 成绩与配置
指标与成绩配置与评测条件证据关系
Accuracy66.7原文 66.7 · 单位 未标注
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
Test subset; 0-shot; video sampled at 1 fps, linearly subsampled to max 1024 frames.
条件
Table 6 caption describes string-match accuracy for multiple-choice VideoQA and LLM-based accuracy for open-ended VideoQA; exact per-benchmark scoring variant is not stated.

直接支持该成绩

在排名中查看此成绩