DocAtlas - Formula Edit Distance: leaderboard
Formula transcription in multilingual document parsing (DocAtlas, Heakl et al., MBZUAI and IBM Research, 2026): 5,862 pages in 82 languages, built by differential rendering of native DOCX files and synthetic right-to-left pages, converted to Markdown and matched to the ground truth with the OmniDocBench pipeline. The score is the normalized edit distance of the LaTeX formulas on a 0-1 scale; lower is better.
Metric: Normalized edit distance of formula LaTeX (0-1) on the 5,862-page DocAtlas benchmark (82 languages), OmniDocBench-style matching. Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 0.43 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=docatlas-formula-edit-distance · How It Works · Data refreshed daily, snapshot 2026-11-12.