Gemini 4 Argon · OSWorld(2.0 offline subset; partial score) 69.2% · 附件证据(ev-g4a-osworld20)

Gemini 4 Argon 的 OSWorld · 2.0 offline subset; partial score 官方出处。Gemini 4 Argon highest thinking · 69.2%。见《Gemini 4 Argon Model evaluation - Approach, methodology & results》PDF p.5, OSWorld-2.0 offline subset partial score row, Gemini 4 Argon column。证据 ev-g4a-osworld20。

原文与定位

Gemini 4 Argon · OSWorld(2.0 offline subset; partial score),Gemini 4 Argon highest thinking配置,Partial score 69.2%(具体成绩公布时间未确认)。见《Gemini 4 Argon Model evaluation - Approach, methodology & results》PDF p.5, OSWorld-2.0 offline subset partial score row, Gemini 4 Argon column。 查看成绩条件

OSWorld-2.0, Offline subset partial score | Gemini 4 Argon 69.2%.
官方材料
Gemini 4 Argon Model evaluation - Approach, methodology & results · PDF 第 5 页
定位
附件 · PDF p.5, OSWorld-2.0 offline subset partial score row, Gemini 4 Argon column
材料版本
5-page PDF; method p.2-4 and results p.5; results as of September 2026; table rendered and visually checked
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 4 Argon · OSWorld(2.0 offline subset; partial score),Gemini 4 Argon highest thinking配置,Partial score 69.2%(具体成绩公布时间未确认)。官方出处 Gemini 4 Argon Model evaluation - Approach, methodology & results(附件,5-page PDF; method p.2-4 and results p.5; results as of September 2026; table rendered and visually checked)PDF p.5, OSWorld-2.0 offline subset partial score row, Gemini 4 Argon column,https://deepmind.google/models/evals-methodology/gemini-4-argon#page=5。Benchlog https://benchlog.lekshome.cc/evidence/gemini-4-argon/ev-g4a-osworld20/

关联成绩与条件

Gemini 4 Argon · OSWorld 成绩与配置
指标与成绩配置与评测条件证据关系
Partial score69.2%原文 69.2% · 单位 %
公布 未确认
Gemini 4 Argon highest thinking
设置
Gemini API model gemini-4-argon; highest thinking settings; default sampling unless benchmark method states otherwise; pass@1 except where stated.
未知项
Exact thinking budget and sampling parameters are not published.、原文仅称 highest thinking settings,未确认对应的公开档位名称。
工具
允许工具
推理预算
未记录
框架
未记录
协议
OSWorld 2.0 offline subset, partial score; maximum over 3 runs, one attempt per run, 1080p, max 500 steps; batched tools and screenshot-only observation.
条件
Uses official OSWorld 2.0 setup and 2026-08-08 patch; distinct from 2.1 82-task offline subset.

直接支持该成绩

在排名中查看此成绩