List your company

Spreadsheet Manipulation on SpreadsheetBench

57.7Pass@1Copilot Agent Mode
7.36420.43233.546.568May 21, 2026

Evaluation results

MethodLinks
Copilot Agent ModeEnvironment Access=Windows, Excel2026.0557.7
ChatGPT agent with .xlsx accessEnvironment Access=OSX, LibreOffice2026.0545.5
Claude Files Opus 4.1Environment Access=Windows, Excel2026.0542.9
ChatGPT agentEnvironment Access=OSX, LibreOffice2026.0535.3
Qwen3-4B-Thinking-2507 + Spreadsheet-RL post-trainingEnvironment Access=Spreadsheet Gym2026.0523.4
OpenAI o3Environment Access=OSX, LibreOffice2026.0523.3
Qwen3-4B-Thinking-2507 + Comprehensive spreadsheet-tool accessEnvironment Access=Spreadsheet Gym2026.0519.3
GPT-4oEnvironment Access=Windows, Excel2026.0518.4
Qwen3-32BEnvironment Access=Spreadsheet Gym2026.0517.6
GPT-4oEnvironment Access=OSX, LibreOffice2026.0516.8
Qwen3-8BEnvironment Access=Spreadsheet Gym2026.0515.9
Qwen3-4B-Thinking-2507 + Spreadsheet-native interaction harnessEnvironment Access=Spreadsheet Gym2026.0515.6
Qwen3-14BEnvironment Access=Spreadsheet Gym2026.0515
Qwen3-4B-Thinking-2507Environment Access=Spreadsheet Gym2026.0512
Qwen3-4BEnvironment Access=Spreadsheet Gym2026.0511
Qwen3-4B-Instruct-2507Environment Access=Spreadsheet Gym2026.059.3