SpreadsheetBench v2 - Debugging Accuracy: leaderboard
SpreadsheetBench v2 evaluates agents completing spreadsheet workflows on real business files. This series reports task accuracy for the debugging category from the official leaderboard, preserving its named agent products and SWE-agent model runs. Accuracy is the percentage of tasks solved, distinct from the paper's cell-match metric. Higher is better.
Metric: Task accuracy (%). Source: spreadsheetbench.github.io. Saturation forecast: Around 2029. 11 models tracked. Updated 2 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | arito | 29 | |
| 2 | WPS AI | 16 | |
| 3 | Claude Opus 4.6 | 12 | #114 |
| 4 | DeepSeek V3.2 | 10 | #417 |
| 5 | GPT-5.2 | 8 | #185 |
| 6 | Gemini 3.1 Pro (Preview) | 7 | #99 |
| 7 | GLM-5 | 7 | #267 |
| 8 | Kimi K2.5 | 4 | #263 |
| 9 | MiniMax-M2.5 | 4 | #589 |
| 10 | Qwen 3.5 397B A17B | 3 | #255 |
| 11 | Decide Agent 2.0 | 0 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=spreadsheetbench-v2-debugging-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-17.