SpreadsheetBench 2
Professional Work
SpreadsheetBench 2 evaluates agents on end-to-end business spreadsheet workflows spanning financial modeling and template completion, debugging, and visualization in complex multi-sheet workbooks. Unlike the isolated manipulations of SpreadsheetBench v1, its 321 tasks require multi-step coordinated operations, cross-sheet reasoning, and deliverable-level outcomes.
Top models (higher is better)
| Model | Score |
|---|---|
| Kimi K3 | 34.8 |
| Claude Fable 5 | 34.7 |
| GPT-5.6 Sol | 32.4 |
| Opus 4.8 | 31.6 |
| GPT-5.5 | 29.1 |
| GLM-5.2 | 28.1 |