All AutosheetBench tasks

Turn a revenue table into a heatmap

FormattingEasyLLM judge scoringLatest run

Passed by all 12 products on their best model.

Prompt and starting spreadsheet

Task files

Prompt

Format table so we can easily observe revenue intensity.

Starting spreadsheet

Data
ABCDEFG
1RegionJanFebMarAprMayJun
2North120014501620175018101940
3South98011201280133014101490
4East132014901570169017601880
5West8709301010110011901250

Expected result

How each AI configuration did

21 of 22 passRun files
Excel and Sheets · All types · All models · All outputs

Pass criteria

All must be met, checked by an LLM judge.
  • A heatmap on B2:G5
  • Low and high revenue clearly differ
  • The headers stay unchanged and uncolored

Passed 21

Autosheet in Excel Opus 5.5 · Medium

Passed10s · 3¢

Autosheet in Excel GPT-6 Astra · Low

Passed8s · 5¢

Autosheet in Sheets Opus 5.5 · Medium

Passed18s · 5¢

Autosheet in Excel Sonnet 5.5 · Medium

Passed6s · 1¢

Claude in Excel Opus 5.5 · High

Passed14s · 12¢

Autosheet in Excel Haiku 5.5 · High

Passed6s · 0.1¢

Claude in Sheets Opus 5.5 · Medium

Passed29s · 11¢

Claude Code + Sheets MCP Opus 5.5 · Medium

Passed46s · N/A

Autosheet in Excel GPT-6.1 Sol · Medium

Passed11s · 7¢

Claude Code + Autosheet MCP Opus 5.5 · Medium

Passed57s · N/A

ChatGPT + Autosheet MCP GPT-6 · Medium

Passed46s · N/A

Autosheet in Sheets Haiku 5.5 · High

Passed19s · 0.1¢

ChatGPT in Excel GPT-6 Astra · Medium

Passed24s · 57¢

Claude in Excel Haiku 5.5 · High

Passed13s · 0.6¢

Claude in Sheets Haiku 5.5 · High

Passed25s · 0.6¢

Copilot in Excel Auto

Passed27s · N/A

ChatGPT + Drive MCP GPT-6 · Medium

Passed44s · N/A

ChatGPT in Excel GPT-6 Luna · High

Passed1m 15s · 0.9¢

Gemini in Sheets Auto

Passed1m 32s · N/A

ChatGPT in Sheets GPT-6 Astra · Medium

Passed48s · 56¢

ChatGPT in Sheets GPT-6 Luna · High

Passed52s · 0.7¢

Failed 1

Autosheet in Excel GPT-6 Luna · High

Failed8s · 0.2¢

Why it failed: Column A width was set to 2.33, truncating the row headers into illegible fragments ('N', 'So', 'Ea', 'W', 'egi') rather than keeping them fully visible.