Claude Sonnet · OSWorld-Verified · Verified benchmark; evaluation protocol noted per source · 历代记录

按发布时间查看 Claude Sonnet 的 2 次发布在 OSWorld-Verified · Verified benchmark; evaluation protocol noted per source 的成绩、配置与披露。

仅为直接前代间的唯一、有效精确值计算差值,要求单位、指标与明确记录的配置和条件一致。

Claude Sonnet 4.6

· 本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)

Claude Sonnet 4.6 · OSWorld-Verified 成绩与配置
指标与成绩配置与评测条件出处
First-attempt success rate72.5%原文 72.5% · 单位 %
公布 未确认
Evaluation settings not fully stated
设置
Only settings explicitly listed in each result's conditions are known.
未知项
Effort, sampling and tool settings are unspecified unless stated in result.conditions.
工具
工具条件未记录
推理预算
未记录
框架
未记录
协议
OSWorld-Verified first-attempt success rate; default settings; 1080p; at most 100 action steps per task; five-run average.
条件
Effort, tools and model sampling settings are not specified in the benchmark section. The main article states scores from Sonnet 4.5 onward use OSWorld-Verified; Sonnet 5 later reports a changed protocol and a different 4.6 comparison value, noted in coverage.md but not imported as a 4.6 result.
在排名中查看此成绩

Claude Sonnet 5

· 本次发布有 1 条成绩记录,含 1 条精确值(可含历史记录)

与直接前代的记录缺少满足差值计算的同条件依据。

Claude Sonnet 5 · OSWorld-Verified 成绩与配置
指标与成绩配置与评测条件出处
First-attempt success rate81.2%原文 81.2% · 单位 %
公布 未确认
Standard configuration: adaptive thinking, max effort
设置
Adaptive thinking at max effort; default temperature and top_p; five-trial averages unless otherwise stated.
未知项
Context-window size is evaluation-dependent; standard configuration is 1M tokens.
工具
工具条件未记录
推理预算
max
框架
未记录
协议
OSWorld-Verified first-attempt success rate, five-run average; 361 tasks and 100 steps per task.
条件
Adaptive thinking at max effort, default 1080p settings. Protocol includes zoom-tool bug fix for batched actions and per-turn token limit increase from 16K to 128K; main page reports only Sonnet 5's score, not the Sonnet 4.6 cross-reference value.
在排名中查看此成绩