SpreadsheetBench v2 - Debugging Cell Match: leaderboard
SpreadsheetBench 2 (2026) evaluates spreadsheet agents on expert-annotated business workflows built from financial reports and corporate filings, run under an SWE-agent-based scaffold. This board is the cell-match score in percent, the average fraction of target cells matching the gold workbook, on tasks that locate and repair injected spreadsheet errors; higher is better.
Metric: Cell-match score (%, Modif.) on the 100 debugging tasks (locating and repairing injected spreadsheet errors) under the SWE-agent-based scaffold (bash, view_xlsx and submit tools, at most 50 turns), thinking mode on, high reasoning effort for GPT-5.2 and Gemini 3.1 Pro. Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 (Thinking) | 50.38 | #102 (Claude Opus 4.6) |
| 2 | Gemini 3.1 Pro (Preview) | 41.86 | #94 |
| 3 | GPT-5.2 (High) | 39.07 | #176 (GPT-5.2) |
| 4 | GLM-5 | 29.06 | #260 |
| 5 | DeepSeek V3.2 (Thinking) | 28.71 | #406 (DeepSeek V3.2) |
| 6 | Qwen 3.5 397B A17B | 24.06 | #246 |
| 7 | Kimi K2.5 | 19.48 | #256 |
| 8 | MiniMax-M2.5 | 16.76 | #582 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=spreadsheetbench-v2-debugging-cell-match · How It Works · Data refreshed daily, snapshot 2026-11-13.