SpreadsheetBench v2 - Cell Match: leaderboard

Metric: Cell-match score (%, Modif.): average fraction of target cells whose recalculated values match the golden workbook, task-weighted over the 297 template, financial-modeling and debugging tasks 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 December 2026. 8 models tracked.

Top models

#ModelScore
1Claude Opus 4.6 (Thinking)77.2
2Gemini 3.1 Pro (Preview)70.91
3GPT-5.2 (High)69.85
4GLM-557.95
5DeepSeek V3.2 (Thinking)54.35
6Qwen 3.5 397B A17B52.96
7Kimi K2.550.31
8MiniMax-M2.542.91

Interactive version: theaggregate.ai/benchmark?slug=spreadsheetbench-v2-cell-match · How It Works · Data refreshed daily, snapshot 2026-09-29.