MPDocBench-Parse: leaderboard
End-to-end parsing of multi-page documents (433 documents, 3,246 pages) across text, formulas, tables, figures, reading order, headings and content that continues across columns or pages; MPDocBench-Parse (Zhou et al., 2026). Overall score from 0 to 100, the mean of the eight task scores with edit distances counted as (1 - NED) x 100; higher is better.
Metric: Overall score (0-100; mean of the eight task scores, with the three edit-distance tasks as (1 - NED) x 100; 433 multi-page documents, 3,246 pages). Source: arxiv.org. Saturation forecast: Estimated already saturated. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B | 73.98 | #477 |
| 2 | Qwen 3.6 Plus | 72.14 | #217 |
| 3 | Gemini 3.1 Pro (Preview) | 71.79 | #91 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mpdocbench-parse · How It Works · Data refreshed daily, snapshot 2026-11-12.