Claude Opus 5.5 · BenchCAD Vision2Code · 1,000-file random subset of 17,900 files · 成绩与配置

Claude Opus 5.5 的 BenchCAD Vision2Code · 1,000-file random subset of 17,900 files 记录。本次发布有 2 条成绩记录,含 2 条精确值(可含历史记录),附 1 条来源定位。

成绩与评测条件

Claude Opus 5.5 · BenchCAD Vision2Code 成绩与配置
指标与成绩配置与评测条件出处
Voxel IoU0.73原文 0.730 voxel IoU without tools · 单位 未标注
公布 未确认
BenchCAD Vision2Code max effort, no tools
设置
Adaptive thinking/max effort; corrected 256 x 256 px per view implementation; random 1,000-file subset; five runs.
未知项
Task-level variance is not separately reported in the card text.
工具
禁止工具
推理预算
max
框架
未记录
协议
BenchCAD Vision2Code; random 1,000-file subset of 17,900; adaptive thinking/max effort; corrected 256 x 256 px per view implementation; five runs; no tools.
条件
Replaces the earlier 128 x 128 px implementation for model comparisons; not expressed as percent accuracy. Source-wide conditions: Corrected intended implementation uses 256 x 256 px per view, four views; random 1,000-file subset; max effort; five runs. Earlier 128 x 128 px implementation is not comparable.
在排名中查看此成绩
Voxel IoU0.962原文 0.962 voxel IoU with tools · 单位 未标注
公布 未确认
BenchCAD Vision2Code max effort, with tools
设置
Adaptive thinking/max effort; container, standard libraries, and image-cropping tool; corrected 256 x 256 px per view implementation; random 1,000-file subset; five runs.
未知项
Task-level variance is not separately reported in the card text.
工具
允许工具
推理预算
max
框架
未记录
协议
BenchCAD Vision2Code; random 1,000-file subset of 17,900; adaptive thinking/max effort; corrected 256 x 256 px per view implementation; container, libraries, and crop tool; five runs.
条件
Replaces the earlier 128 x 128 px implementation for model comparisons; not expressed as percent accuracy. Source-wide conditions: Corrected intended implementation uses 256 x 256 px per view, four views; random 1,000-file subset; max effort; five runs. Earlier 128 x 128 px implementation is not comparable.
在排名中查看此成绩

出处与披露

测试版本与指标

别名 BenchCAD Vision2Code

  • Voxel IoU · 单位未标注 · 越高越好