PureDocBench - Clean - Formula CDM: leaderboard

Formula recognition on the clean track of PureDocBench (2026), a document-parsing benchmark of 1,475 source-traceable pages (pages rendered straight from their HTML/CSS source), scored by Character Detection Matching of rendered formulas from 0 to 100. Higher is better.

Metric: Formula Character Detection Matching score (0-100) between the rendered predicted and reference LaTeX formulas, on the clean track (pages rendered straight from their HTML/CSS source) of 1,475 source pages in 10 domains and 66 subcategories, rendered from LLM-written HTML/CSS with source-linked annotations for text, formulas, tables and reading order; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 58 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1TeleOCR82.33
2GLM-5.3 Flash80.16#137
3Gemini 3.6 Flash78.29#84
4Claude Opus 575.92#26
5GPT-5.6 Sol74.96#36
6Qwen3.8-Flash-Next72.28#122
7Qwen 3.8 27B71.68#232
8WeVisDoc-2B71.25
9Qwen 3.6 35B A3B70.39#485
10Qwen 3.5 4B69.96#1270
11WeVisDoc-4B69.95
12OvisOCR269.66
13Qwen 3.6 27B69.36#358
14FD-RL68.21
15Qwen 3.5 122B A10B67.96#350

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=puredocbench-clean-formula-cdm · How It Works · Data refreshed daily, snapshot 2026-11-22.