VDE Bench (Global) - BLEU-4: leaderboard
Metric: BLEU-4 of the OCR text against the gold text over matched blocks (0-1, shown times 100), over the full page, every edited region counted; VDE Bench's 942 human-verified instruction-based edits of dense Chinese and English document pages (papers, slides, posters, exams, newspapers; text replacement, insertion and deletion and table edits; gold edited pages verified by 15 annotators); the edited page is read with PaddleOCR-VL and compared with the gold page, default generation settings; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen-Image-Edit (VDE Bench checkpoint unspecified) | 43 | |
| 2 | FireRed-Image-Edit-1.0 | 38.8 | |
| 3 | Step1X-Edit (VDE Bench checkpoint unspecified) | 37.5 | |
| 4 | LongCat-Image-Edit | 37.5 | |
| 5 | InstructPix2Pix | 13.2 | |
| 6 | ICEdit | 10 |
Interactive version: theaggregate.ai/benchmark?slug=vde-bench-global-bleu-4 · How It Works · Data refreshed daily, snapshot 2026-10-11.