Who leads the olmOCR-Bench benchmark?

Chandra v0.1.0 currently leads olmOCR-Bench with a score of 99.9 on Base.

What is the state-of-the-art score on olmOCR-Bench?

The state-of-the-art result on olmOCR-Bench is 99.9 (Base), achieved by Chandra v0.1.0 as of 2026.

How many models are tracked on olmOCR-Bench?

Codesota tracks 32 models on olmOCR-Bench across 9 metrics.

When was the olmOCR-Bench leaderboard last updated?

The olmOCR-Bench leaderboard on Codesota includes results through 2026, with the earliest tracked result from 2025.

Codesota · Benchmark · olmOCR-BenchHome/Leaderboards/Vision & Documents/Document Parsing/olmOCR-Bench

Allen Institute for AI

olmOCR-Bench.

Name: olmOCR-Bench Benchmark Results
Creator: Allen Institute for AI
Published: 2025-01-01
License: https://creativecommons.org/licenses/by/4.0/

7,010 unit tests across 1,402 PDF documents. Tests parsing of tables, math, multi-column layouts, old scans, and more.

Paper ↗Leaderboard ↓Lineage

§ 01 · SOTA history

Year over year.

§ 02 · Leaderboard

Results by metric.

Found a wrong score or missing run?

Use row edits to send a sourced correction into moderation.

Add / edit result ↗Report issue ↗

Base

Base is the reported evaluation metric for olmOCR-Bench. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Baseverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Edit
01	Chandra v0.1.0 Base clean document parsing. Near-perfect	unverified	99.9	2025	Source ↗	Edit result
02	chandra-ocr-0.1.0 Base clean document parsing. Near-perfect	paper	99.9	2025	Source ↗	Edit result
03	olmOCR v0.4.0 olmOCR 2. Sub-category: base clean documents.	paper	99.7	2025	Source ↗	Edit result
04	olmocr-v0.4.0 olmOCR 2. Base clean documents sub-category.	paper	99.7	2025	Source ↗	Edit result
05	LightOnOCR-2-1B LightOnOCR-2-1B. Base clean documents sub-category.	paper	99.6	2026	Source ↗	Edit result
06	Qianfan-OCR Qianfan-OCR. Base clean documents sub-category.	paper	99.6	2026	Source ↗	Edit result

Headers Footers

Headers Footers is the reported evaluation metric for olmOCR-Bench. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Headers Footersverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Edit
01	olmOCR v0.4.0 olmOCR 2. Sub-category: headers/footers.	paper	96.1	2025	Source ↗	Edit result
02	olmocr-v0.4.0 olmOCR 2. Headers/footers sub-category.	paper	96.1	2025	Source ↗	Edit result
03	olmOCR v0.3.0 #1 on headers/footers extraction	unverified	95.1	2025	Source ↗	Edit result
04	olmocr-v0.3.0 #1 on headers/footers extraction	paper	95.1	2025	Source ↗	Edit result
05	chandra-ocr-0.1.0 Header/footer extraction	paper	90.8	2025	Source ↗	Edit result
06	Chandra v0.1.0 Header/footer extraction	unverified	90.8	2025	Source ↗	Edit result

Long Tiny Text

Long Tiny Text is the reported evaluation metric for olmOCR-Bench. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Long Tiny Textverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Edit
01	Chandra v0.1.0 Long documents with tiny text. #1 in category	unverified	92.3	2025	Source ↗	Edit result
02	chandra-ocr-0.1.0 Long documents with tiny text. #1 in category	paper	92.3	2025	Source ↗	Edit result
03	LightOnOCR-2-1B LightOnOCR-2-1B. Long tiny text sub-category.	paper	91.4	2026	Source ↗	Edit result
04	olmocr-v0.4.0 olmOCR 2. Long tiny text sub-category.	paper	81.9	2025	Source ↗	Edit result
05	olmOCR v0.4.0 olmOCR 2. Sub-category: long tiny text.	paper	81.9	2025	Source ↗	Edit result
06	Qianfan-OCR Qianfan-OCR. Long tiny text sub-category.	paper	80.4	2026	Source ↗	Edit result

Multi Column

Multi Column is the reported evaluation metric for olmOCR-Bench. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Multi Columnverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Edit
01	Qianfan-OCR Qianfan-OCR. Multi-column layout sub-category.	paper	92.2	2026	Source ↗	Edit result
02	LightOnOCR-2-1B LightOnOCR-2-1B. Multi-column layout sub-category.	paper	84.8	2026	Source ↗	Edit result
03	olmocr-v0.4.0 olmOCR 2. Multi-column layout sub-category.	paper	83.7	2025	Source ↗	Edit result
04	olmOCR v0.4.0 olmOCR 2. Sub-category: multi-column layout.	paper	83.7	2025	Source ↗	Edit result
05	Chandra v0.1.0 Multi-column document parsing	unverified	81.2	2025	Source ↗	Edit result
06	chandra-ocr-0.1.0 Multi-column document parsing	paper	81.2	2025	Source ↗	Edit result

Arxiv

Arxiv is the reported evaluation metric for olmOCR-Bench. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Arxivverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Edit
01	LightOnOCR-2-1B LightOnOCR-2-1B. ArXiv math documents sub-category.	paper	89.6	2026	Source ↗	Edit result
02	marker-1.10.0 #1 on ArXiv paper parsing	paper	83.8	2025	Source ↗	Edit result
03	Marker 1.10.0 #1 on ArXiv paper parsing	unverified	83.8	2025	Source ↗	Edit result
04	olmOCR v0.4.0 olmOCR 2 (arxiv:2510.19817). Sub-category: ArXiv math documents.	paper	83	2025	Source ↗	Edit result
05	olmocr-v0.4.0 olmOCR 2 (arxiv:2510.19817). ArXiv math documents sub-category.	paper	83	2025	Source ↗	Edit result
06	chandra-ocr-0.1.0 ArXiv paper parsing. Marker leads (83.8)	paper	82.2	2025	Source ↗	Edit result
07	Chandra v0.1.0 ArXiv paper parsing. Marker leads (83.8)	unverified	82.2	2025	Source ↗	Edit result
08	Qianfan-OCR Qianfan-OCR. ArXiv math documents sub-category.	paper	80.1	2026	Source ↗	Edit result

Tables

Tables is the reported evaluation metric for olmOCR-Bench. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Tablesverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Edit
01	LightOnOCR-2-1B LightOnOCR-2-1B. Table recognition sub-category.	paper	89	2026	Source ↗	Edit result
02	dots.ocr 3B #1 on table recognition	unverified	88.3	2025	Source ↗	Edit result
03	dots-ocr-3b #1 on table recognition	paper	88.3	2025	Source ↗	Edit result
04	Chandra v0.1.0 Table recognition category. Near-best (dots.ocr: 88.3)	unverified	88	2025	Source ↗	Edit result
05	chandra-ocr-0.1.0 Table recognition category. Near-best (dots.ocr: 88.3)	paper	88	2025	Source ↗	Edit result
06	olmocr-v0.4.0 olmOCR 2. Table recognition sub-category.	paper	84.9	2025	Source ↗	Edit result
07	olmOCR v0.4.0 olmOCR 2. Sub-category: table recognition.	paper	84.9	2025	Source ↗	Edit result
08	Qianfan-OCR Qianfan-OCR. Table recognition sub-category.	paper	81.6	2026	Source ↗	Edit result

Accuracy

Accuracy is the reported evaluation metric for olmOCR-Bench. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Accuracyverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Edit
01	Infinity-Parser2-Pro	unverified	87.6	2026	Paper ↗	Edit result
02	Chandra 2	unverified	85.9	2026	Paper ↗Code ↗	Edit result
03	dots.mocr	unverified	83.9	2026	Paper ↗Code ↗	Edit result
04	LightOnOCR-2-1B	unverified	83.2	2026	Paper ↗Source ↗	Edit result
05	Chandra	unverified	83.1	2025	Paper ↗	Edit result
06	Infinity-Parser 7B	unverified	82.5	2025	Paper ↗Code ↗	Edit result
07	olmOCR-2-7B-1025 (7B)	unverified	82.4	2025	Paper ↗	Edit result
08	Falcon-OCR	unverified	80.3	2026	Paper ↗Code ↗	Edit result
09	PaddleOCR-VL	unverified	80	2025	Paper ↗Code ↗	Edit result
10	Qianfan-OCR	unverified	79.8	2026	Paper ↗Code ↗	Edit result
11	dots.ocr	unverified	79.1	2025	Paper ↗Code ↗	Edit result
12	MinerU2.5	unverified	77.5	2025	Paper ↗Code ↗	Edit result
13	DeepSeek-OCR-2	unverified	76.3	2026	Paper ↗Code ↗	Edit result
14	LightOnOCR-1B-1025	unverified	76.1	2026	Paper ↗	Edit result

Old Scans Math

Old Scans Math is the reported evaluation metric for olmOCR-Bench. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Old Scans Mathverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Edit
01	LightOnOCR-2-1B LightOnOCR-2-1B. Old scans with math sub-category.	paper	85.6	2026	Source ↗	Edit result
02	olmocr-v0.4.0 olmOCR 2. Old scans with math sub-category.	paper	82.3	2025	Source ↗	Edit result
03	olmOCR v0.4.0 olmOCR 2. Sub-category: old scans with math.	paper	82.3	2025	Source ↗	Edit result
04	chandra-ocr-0.1.0 Mathematical notation in old scans. #1, leads by 5.4 points	paper	80.3	2025	Source ↗	Edit result
05	Chandra v0.1.0 Mathematical notation in old scans. #1, leads by 5.4 points	unverified	80.3	2025	Source ↗	Edit result
06	olmocr-v0.3.0 #2 on math in old scans	paper	79.9	2025	Source ↗	Edit result
07	olmOCR v0.3.0 #2 on math in old scans	unverified	79.9	2025	Source ↗	Edit result

Pass Rate

Pass Rate is the reported evaluation metric for olmOCR-Bench. Codesota tracks published model scores on this metric so readers can compare state-of-the-art results across sources and model families.

Higher is better

Trust tiers for Pass Rateverifiedpapervendorcommunityunverified

Rank	Model	Trust	Score	Year	Links	Edit
01	dots.mocr dots.mocr (arxiv:2603.13032). Current SOTA on olmOCR-Bench (83.9 ± 0.9). 3B multimodal model.	unverified	83.9	2026	Source ↗	Edit result
02	LightOnOCR-2-1B LightOnOCR-2-1B. SOTA at publication (Jan 2026). 9x smaller than Chandra-9B, 3.3x faster.	paper	83.2	2026	Source ↗	Edit result
03	chandra-ocr-0.1.0 7,010 unit tests across 1,402 PDF documents. #1 overall on olmOCR-Bench.	paper	83.1	2025	Source ↗	Edit result
04	Chandra v0.1.0 7,010 unit tests across 1,402 PDF documents. #1 overall on olmOCR-Bench.	unverified	83.1	2025	Source ↗	Edit result
05	infinity-parser-7b	paper	82.5	2025	Source ↗	Edit result
06	Infinity-Parser 7B	unverified	82.5	2025	Source ↗	Edit result
07	olmOCR v0.4.0	unverified	82.4	2025	Source ↗	Edit result
08	olmocr-v0.4.0	paper	82.4	2025	Source ↗	Edit result
09	paddleocr-vl	paper	80	2025	Source ↗	Edit result
10	Qianfan-OCR Baidu Qianfan-OCR 4B (Qwen3-4B + Qianfan-ViT), Apache 2.0, 192 langs. Layout-as-Thought.	paper	79.8	2026	Source ↗	Edit result
11	Qwen3-VL-4B Qwen3-VL-4B score from Qianfan-OCR Table 3. General-purpose VLM baseline.	paper	79.2	2026	Source ↗	Edit result
12	PaddleOCR-VL-1.5 PaddleOCR-VL-1.5 (arxiv:2601.21957). Score from Qianfan-OCR comparison table (Table 3). 0.9B params.	paper	79.1	2026	Source ↗	Edit result
13	dots.ocr 3B	unverified	79.1	2025	Source ↗	Edit result
14	dots-ocr-3b	paper	79.1	2025	Source ↗	Edit result
15	mistral-ocr-3 Estimated based on 74% win rate vs OCR 2	paper	78	2025	Source ↗	Edit result
16	Mistral OCR 3 Estimated based on 74% win rate vs OCR 2	unverified	78	2025	Source ↗	Edit result
17	Marker 1.10.0	unverified	76.5	2025	Source ↗	Edit result
18	marker-1.10.0	paper	76.5	2025	Source ↗	Edit result
19	marker-1.10.1	paper	76.1	2025	Source ↗	Edit result
20	Marker 1.10.1	unverified	76.1	2025	Source ↗	Edit result