Spreadsheet Manipulation and Reasoning on SpreadsheetBench
28.8Soft AccuracyTabClaw
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TabClawCategory=Ours, Backbone=DeepSeek-v3.22026.06 | 28.8 | 24.5 | |
| BinderCategory=Context Augmentation, Backbone=DeepSeek-v3.22026.06 | 24.1 | 18.9 | |
| SpreadsheetLLMCategory=Context Augmentation, Backbone=DeepSeek-v3.22026.06 | 23.3 | 19.8 | |
| SheetAgentCategory=Autonomous Agent, Backbone=DeepSeek-v3.22026.06 | 22.3 | 18.5 | |
| VanillaCategory=Direct Inference, Backbone=DeepSeek-v3.22026.06 | 18.8 | 15.5 | |
| ReActCategory=Autonomous Agent, Backbone=DeepSeek-v3.22026.06 | 18.8 | 16.1 | |
| Chain-of-TableCategory=Context Augmentation, Backbone=DeepSeek-v3.22026.06 | 14.8 | 12.9 | |
| NormTabCategory=Context Augmentation, Backbone=DeepSeek-v3.22026.06 | 14.8 | 12.2 |