Model reports

AutomationBench

Published by Zapier over its own automation surface, and cards report different task subsets of it.

评测信息与证据

本站保留来源报告的结果，不代表独立复测。请先核对任务版本、评测协议、分数单位和报告日期；不同运行条件不能直接合并为统一排名。

按来源中的观察顺序展示，顺序不代表排名。日期保留原始精度。

模型	数值	单位	协议	报告日期与口径	来源
Claude Opus 5	26.0	percent	max effort, private held-out eval set	2026-07-24 · document_publication	Claude Opus 5
DeepSeek-V4-Pro-0813	31.8	percent	max reasoning effort, temp 1.0, top_p 0.95	2026-08-13 · document_publication	DeepSeek-V4-Pro-0813
Kimi K3	30.8	percent	reasoning=max	2026-06-13 · document_publication	Kimi K3
Hy4 preview	32.1	percent	AutomationBench v1.0.6, as reported	2026-08-28 · document_publication	Hy4 preview
GLM-5.3-Flash	48.8	percent	AutomationBench v1.0.6, with null-type handling fix (PR	2026-08-27 · document_publication	GLM-5.3-Flash