DragOn - Cell Selection: leaderboard

Metric: Success rate (%) on the cell-selection domain (drag over a spreadsheet cell range in a LibreOffice viewport) of the private 1,000-example DragOn test split (250 per domain): from one screenshot and instruction the model outputs the drag's start and end points as JSON, and both must fall inside the pixel-aligned source and target boxes; each model at its documented settings (Claude extended thinking off, Kimi-K2.5 reasoning off, local Qwen and Holo models with chain-of-thought on); higher is better. Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.

Top models

#ModelScore
1Claude Opus 4.737.2
2GPT-5.428
3Qwen 3.5 397B A17B16.4
4Kimi K2.5 (Non-reasoning)14
5Qwen 3.5 35B A3B0.4
6Claude Sonnet 4.50

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