AI Radar
Top stories
Importance 2 of 5Covered by 1 source

GDP.xlsx benchmark tests frontier models on professional spreadsheet tasks

A new benchmark offers 70 realistic workplace requests across 13 domains, each with expert rubrics and real workbooks totaling 2.18 million cells. Agents use spreadsheet tools and an agentic judge grades results.

Why it matters

It gives a more realistic gauge of how well AI agents handle everyday office spreadsheet work.