Documents & OCR

OCR and document-understanding models, the apps that read and answer questions about your files, and parsing libraries and APIs that turn PDFs and scans into structured text, locally or in the cloud.

Models are ranked by OmniDocBench v1.6 (end-to-end, v1.6_full), ParseBench (models), MDPBench (multilingual document parsing), then OCR Arena for models without a result on the first. Tools with their own result on OmniDocBench v1.6 (end-to-end, v1.6_full): pipeline tools are ranked by it (each board ranks one kind of product); other tools inherit the tier of the best model they offer. How we rank.

Best overallSTeleOCR97 on OmniDocBench v1.6 (end-to-end, v1.6_full)
Best valueNo qualifying option yet
Best free startNo qualifying option yet
Best local (≤24 GB GPU)STeleOCR~5 GB VRAM · 97 on OmniDocBench v1.6 (end-to-end, v1.6_full)

Tier: how close to the best result on the task's main public benchmark
Outlined: a tool inherits the tier of the best model it offers
No public result yet
Cloud · Local: runs on your own hardware

Models

The engines themselves. Use them through an API or inside the apps on the right.

Cloud

–

Claude Fable 5.1Anthropic · $20/M tokensNo public result yet –

Claude Haiku 5.5Anthropic · $0.2/M tokensNo public result yet –

Claude Opus 5.5Anthropic · $8/M tokensNo public result yet –

Claude Sonnet 5.5Anthropic · $4/M tokensNo public result yet –

Gemini 3.1 Pro (Preview)Google · $4.5/M tokensNo public result yet –

Gemini 3.8 FlashGoogle · $1.5/M tokensNo public result yet –

GPT-6 AstraOpenAI · $20/M tokensNo public result yet –

GPT-6 LunaOpenAI · $0.2/M tokensNo public result yet –

GPT-6.1 SolOpenAI · $4/M tokensNo public result yet –

Mistral OCR 4.1Mistral AINo public result yet

Local

TeleOCRXingChen-AGI · ~5 GB VRAM97 on OmniDocBench v1.6 (end-to-end, v1.6_full)$S

OvisOCR2ATH-MaaS · ~4 GB VRAM96 on OmniDocBench v1.6 (end-to-end, v1.6_full)$S

PaddleOCR-VL-1.6Baidu (PaddlePaddle) · ~4 GB VRAM96 on OmniDocBench v1.6 (end-to-end, v1.6_full)S

MinerU2.5-Pro-2605-1.2BOpenDataLab (Shanghai AI Laboratory) · ~4 GB VRAM96 on OmniDocBench v1.6 (end-to-end, v1.6_full)S

GLM-OCRZ.ai · ~5 GB VRAM95 on OmniDocBench v1.6 (end-to-end, v1.6_full)S

Unlimited-OCRBaidu · ~8 GB VRAM94 on OmniDocBench v1.6 (end-to-end, v1.6_full)S

Qianfan-OCRBaidu · ~11 GB VRAM94 on OmniDocBench v1.6 (end-to-end, v1.6_full)S

Youtu-ParsingTencent (Youtu) · ~7 GB VRAM94 on OmniDocBench v1.6 (end-to-end, v1.6_full)S

FireRed-OCRFireRed Team · ~6 GB VRAM93 on OmniDocBench v1.6 (end-to-end, v1.6_full)A

DeepSeek-OCR 2DeepSeek · ~9 GB VRAM90 on OmniDocBench v1.6 (end-to-end, v1.6_full)A

Dolphin-v2ByteDance · ~9 GB VRAM90 on OmniDocBench v1.6 (end-to-end, v1.6_full)–

Chandra OCR 2Datalab · ~13 GB VRAMNo public result yet –

dots.mocrrednote hi lab (dots studio) · ~8 GB VRAMNo public result yet –

Falcon-OCRTechnology Innovation Institute (TII) · ~2 GB VRAMNo public result yet –

Gemma 4 (31B)Google DeepMind · ~19 GB VRAMNo public result yet –

GLM-5.3-FlashZ.ai · ~185 GB VRAMNo public result yet –

Granite Vision 4.1 4BIBM · ~10 GB VRAMNo public result yet –

granite-docling-258MIBM · ~3 GB VRAMNo public result yet –

HPD-ParsingBaidu (PaddlePaddle) · ~4 GB VRAMNo public result yet –

HunyuanOCR-1.5Tencent (Hunyuan) · ~4 GB VRAMNo public result yet –

Infinity-Parser2-Flashinfly · ~6 GB VRAMNo public result yet –

Infinity-Parser2-Proinfly · ~72 GB VRAMNo public result yet –

jina-ocr-v1Jina AI · ~8 GB VRAMNo public result yet –

Kimi K3Moonshot AI · ~1600 GB VRAMNo public result yet –

LightOnOCR-3 (4B, 1B, 0.8B)LightOn · ~11 GB VRAMNo public result yet –

Logics-Parsing-V3Alibaba (Logics-MLLM) · ~3 GB VRAMNo public result yet –

MonkeyOCRv2-B-Parsingzenosai (MonkeyOCR team) · ~3 GB VRAMNo public result yet –

Nanonets-OCR2-3BNanonets · ~9 GB VRAMNo public result yet –

olmOCR-2-7B-1025Allen Institute for AI (Ai2) · ~12 GB VRAMNo public result yet –

Qwen3.8-27BAlibaba (Qwen) · ~17 GB VRAMNo public result yet –

Surya OCR 2Datalab · ~3 GB VRAMNo public result yet –

WeVisDoc (2B and 4B)Tencent · ~10 GB VRAMNo public result yet

Apps

Ready-made apps with a UI: sign in or install, then start.

Cloud

–

Adobe Acrobat (AI Assistant)Adobe · $9.99/moNo public result yet –

ChatGPTOpenAI · Free tier · $20/moNo public result yet –

ChatPDFChatPDF · Free tierNo public result yet –

ClaudeAnthropic · Free tier · $20/moNo public result yet –

Foxit PDF EditorFoxit · $11/moNo public result yet –

GeminiGoogle · Free tier · $4.99/moNo public result yet –

Gemini Notebook (formerly NotebookLM)Google · Free tierNo public result yet –

HumataHumata · Free tier · $9.99/moNo public result yet –

MathpixMathpix · Free tier · $4.99/moNo public result yet –

PDFgearPDFgear · FreeNo public result yet –

SmallpdfSmallpdf · Free tierNo public result yet

Local

–

ABBYY FineReader PDFABBYYNo public result yet

Parsers & APIs

Libraries and APIs that convert PDFs and scans into Markdown, tables or JSON for your own pipelines.

Cloud

–

Amazon TextractAmazon Web Services · Free tierNo public result yet –

anyformatanyformat · Free tierNo public result yet –

Azure Document IntelligenceMicrosoft · Free tierNo public result yet –

Document AIGoogle Cloud · Free tierNo public result yet –

ExtendExtend · Free tier · $500/moNo public result yet –

LandingAI Agentic Document ExtractionLandingAI · Free tier · $250/moNo public result yet –

LlamaParseLlamaIndex · Free tier · $50/moNo public result yet –

MathpixMathpix · Free tier · $4.99/moNo public result yet –

Nutrient Document Web ServicesNutrient · Free tierNo public result yet –

PulsePulse AI · Free tierNo public result yet –

ReductoReducto · Free tierNo public result yet –

Upstage Document ParseUpstage · Free tierNo public result yet

Local

PaddleOCRBaidu (PaddlePaddle) · Freevia PaddleOCR-VL-1.6$S MinerUOpenDataLab (Shanghai AI Laboratory) · Freevia MinerU2.5-Pro-2605-1.2B –

MarkItDownMicrosoft · FreeNo public result yet –

Tesseract OCRTesseract OCR project · FreeNo public result yet – DoclingDocling Project (LF AI & Data) · FreeNo public result yet –

MarkerDatalab · FreeNo public result yet –

OCRmyPDFOCRmyPDF project · FreeNo public result yet –

OpenDataLoader PDFOpenDataLoader project · FreeNo public result yet –

UnstructuredUnstructured · FreeNo public result yet –

LiteParseLlamaIndex · FreeNo public result yet –

ChunkrLumina AI · FreeNo public result yet –

PyMuPDF4LLMArtifex · FreeNo public result yet

Quality vs. setup

The same tools as the map, placed by how good the results can be (up) and how much setup they take (right). Top-left is the easiest way to top quality; further right buys control, privacy or price at the cost of setup. Models are left out: you reach them through these tools.

Cloud
Local: runs on your hardware
Hollow: tier inherited from the best model the tool offers

Not placed (no public result yet): MarkItDown, Tesseract OCR, Docling, Marker, OCRmyPDF, OpenDataLoader PDF, Unstructured, LiteParse, Chunkr, PyMuPDF4LLM, ABBYY FineReader PDF, Adobe Acrobat (AI Assistant), Amazon Textract, anyformat, Azure Document Intelligence, ChatGPT, ChatPDF, Claude, Document AI, Extend, Foxit PDF Editor, Gemini, Gemini Notebook (formerly NotebookLM), Humata, LandingAI Agentic Document Extraction, LlamaParse, Mathpix, Nutrient Document Web Services, PDFgear, Pulse, Reducto, Smallpdf, Upstage Document Parse

Everything, in order

All documents & ocr options in quality order
Tier	Name	Approach	Runs	Rated by	Price	Skill
S	TeleOCR XingChen-AGI	Models	Local	97 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~5 GB VRAM	—
S	OvisOCR2 ATH-MaaS	Models	Local	96 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~4 GB VRAM	—
S	PaddleOCR-VL-1.6 Baidu (PaddlePaddle)	Models	Local	96 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~4 GB VRAM	—
S	MinerU2.5-Pro-2605-1.2B OpenDataLab (Shanghai AI Laboratory)	Models	Local	96 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~4 GB VRAM	—
S	GLM-OCR Z.ai	Models	Local	95 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~5 GB VRAM	—
S	Unlimited-OCR Baidu	Models	Local	94 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~8 GB VRAM	—
S	Qianfan-OCR Baidu	Models	Local	94 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~11 GB VRAM	—
S	Youtu-Parsing Tencent (Youtu)	Models	Local	94 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~7 GB VRAM	—
S	FireRed-OCR FireRed Team	Models	Local	93 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~6 GB VRAM	—
S	PaddleOCR Baidu (PaddlePaddle)	Parsers & APIs	Local	via PaddleOCR-VL-1.6	Free	advanced
S	MinerU OpenDataLab (Shanghai AI Laboratory)	Parsers & APIs	Local	via MinerU2.5-Pro-2605-1.2B Cloud models: tier S	Free	intermediate
A	DeepSeek-OCR 2 DeepSeek	Models	Local	90 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~9 GB VRAM	—
A	Dolphin-v2 ByteDance	Models	Local	90 on OmniDocBench v1.6 (end-to-end, v1.6_full)	~9 GB VRAM	—
–	MarkItDown Microsoft	Parsers & APIs	Local	No public result yet	Free	advanced
–	Tesseract OCR Tesseract OCR project	Parsers & APIs	Local	No public result yet	Free	advanced
–	Docling Docling Project (LF AI & Data)	Parsers & APIs	Local	No public result yet	Free	advanced
–	Marker Datalab	Parsers & APIs	Local	No public result yet Cloud models: tier A	Free	advanced
–	OCRmyPDF OCRmyPDF project	Parsers & APIs	Local	No public result yet	Free	advanced
–	OpenDataLoader PDF OpenDataLoader project	Parsers & APIs	Local	No public result yet	Free	advanced
–	Unstructured Unstructured	Parsers & APIs	Local	No public result yet	Free	advanced
–	LiteParse LlamaIndex	Parsers & APIs	Local	No public result yet	Free	advanced
–	Chunkr Lumina AI	Parsers & APIs	Local	No public result yet	Free	advanced
–	PyMuPDF4LLM Artifex	Parsers & APIs	Local	No public result yet	Free	advanced
–	ABBYY FineReader PDF ABBYY	Apps	Local	No public result yet	—	beginner
–	Adobe Acrobat (AI Assistant) Adobe	Apps	Cloud	No public result yet	$9.99/mo	beginner
–	Amazon Textract Amazon Web Services	Parsers & APIs	Cloud	No public result yet	Free tier	intermediate
–	anyformat anyformat	Parsers & APIs	Cloud	No public result yet	Free tier	intermediate
–	Azure Document Intelligence Microsoft	Parsers & APIs	Cloud	No public result yet	Free tier	intermediate
–	Chandra OCR 2 Datalab	Models	Local	No public result yet	~13 GB VRAM	—
–	ChatGPT OpenAI	Apps	Cloud	No public result yet	Free tier · $20/mo	beginner
–	ChatPDF ChatPDF	Apps	Cloud	No public result yet	Free tier	beginner
–	Claude Anthropic	Apps	Cloud	No public result yet	Free tier · $20/mo	beginner
–	Claude Fable 5.1 Anthropic	Models	Cloud	No public result yet	$20/M tokens	—
–	Claude Haiku 5.5 Anthropic	Models	Cloud	No public result yet	$0.2/M tokens	—
–	Claude Opus 5.5 Anthropic	Models	Cloud	No public result yet	$8/M tokens	—
–	Claude Sonnet 5.5 Anthropic	Models	Cloud	No public result yet	$4/M tokens	—
–	Document AI Google Cloud	Parsers & APIs	Cloud	No public result yet	Free tier	intermediate
–	dots.mocr rednote hi lab (dots studio)	Models	Local	No public result yet	~8 GB VRAM	—
–	Extend Extend	Parsers & APIs	Cloud	No public result yet	Free tier · $500/mo	intermediate
–	Falcon-OCR Technology Innovation Institute (TII)	Models	Local	No public result yet	~2 GB VRAM	—
–	Foxit PDF Editor Foxit	Apps	Cloud	No public result yet	$11/mo	beginner
–	Gemini Google	Apps	Cloud	No public result yet	Free tier · $4.99/mo	beginner
–	Gemini 3.1 Pro (Preview) Google	Models	Cloud	No public result yet	$4.5/M tokens	—
–	Gemini 3.8 Flash Google	Models	Cloud	No public result yet	$1.5/M tokens	—
–	Gemini Notebook (formerly NotebookLM) Google	Apps	Cloud	No public result yet	Free tier	beginner
–	Gemma 4 (31B) Google DeepMind	Models	Local	No public result yet	~19 GB VRAM	—
–	GLM-5.3-Flash Z.ai	Models	Local	No public result yet	~185 GB VRAM	—
–	GPT-6 Astra OpenAI	Models	Cloud	No public result yet	$20/M tokens	—
–	GPT-6 Luna OpenAI	Models	Cloud	No public result yet	$0.2/M tokens	—
–	GPT-6.1 Sol OpenAI	Models	Cloud	No public result yet	$4/M tokens	—
–	Granite Vision 4.1 4B IBM	Models	Local	No public result yet	~10 GB VRAM	—
–	granite-docling-258M IBM	Models	Local	No public result yet	~3 GB VRAM	—
–	HPD-Parsing Baidu (PaddlePaddle)	Models	Local	No public result yet	~4 GB VRAM	—
–	Humata Humata	Apps	Cloud	No public result yet	Free tier · $9.99/mo	beginner
–	HunyuanOCR-1.5 Tencent (Hunyuan)	Models	Local	No public result yet	~4 GB VRAM	—
–	Infinity-Parser2-Flash infly	Models	Local	No public result yet	~6 GB VRAM	—
–	Infinity-Parser2-Pro infly	Models	Local	No public result yet	~72 GB VRAM	—
–	jina-ocr-v1 Jina AI	Models	Local	No public result yet	~8 GB VRAM	—
–	Kimi K3 Moonshot AI	Models	Local	No public result yet	~1600 GB VRAM	—
–	LandingAI Agentic Document Extraction LandingAI	Parsers & APIs	Cloud	No public result yet	Free tier · $250/mo	intermediate
–	LightOnOCR-3 (4B, 1B, 0.8B) LightOn	Models	Local	No public result yet	~11 GB VRAM	—
–	LlamaParse LlamaIndex	Parsers & APIs	Cloud	No public result yet	Free tier · $50/mo	intermediate
–	Logics-Parsing-V3 Alibaba (Logics-MLLM)	Models	Local	No public result yet	~3 GB VRAM	—
–	Mathpix Mathpix	Apps + Parsers & APIs	Cloud	No public result yet	Free tier · $4.99/mo	beginner
–	Mistral OCR 4.1 Mistral AI	Models	Cloud	No public result yet	—	—
–	MonkeyOCRv2-B-Parsing zenosai (MonkeyOCR team)	Models	Local	No public result yet	~3 GB VRAM	—
–	Nanonets-OCR2-3B Nanonets	Models	Local	No public result yet	~9 GB VRAM	—
–	Nutrient Document Web Services Nutrient	Parsers & APIs	Cloud	No public result yet	Free tier	intermediate
–	olmOCR-2-7B-1025 Allen Institute for AI (Ai2)	Models	Local	No public result yet	~12 GB VRAM	—
–	PDFgear PDFgear	Apps	Cloud	No public result yet	Free	beginner
–	Pulse Pulse AI	Parsers & APIs	Cloud	No public result yet	Free tier	intermediate
–	Qwen3.8-27B Alibaba (Qwen)	Models	Local	No public result yet	~17 GB VRAM	—
–	Reducto Reducto	Parsers & APIs	Cloud	No public result yet	Free tier	intermediate
–	Smallpdf Smallpdf	Apps	Cloud	No public result yet	Free tier	beginner
–	Surya OCR 2 Datalab	Models	Local	No public result yet	~3 GB VRAM	—
–	Upstage Document Parse Upstage	Parsers & APIs	Cloud	No public result yet	Free tier	intermediate
–	WeVisDoc (2B and 4B) Tencent	Models	Local	No public result yet	~10 GB VRAM	—

Benchmarks used

OmniDocBench v1.6 (end-to-end, v1.6_full) by OpenDataLab (Shanghai AI Laboratory) (automated tests, Overall page-parsing score: ((1 − text edit distance) × 100 + table TEDS + formula CDM) / 3) · snapshot 11 Sep 2026
End-to-end table in the GitHub README (caption 'OmniDocBench (v1.6_full)'); last leaderboard update 2026-09-11 (TeleOCR, OvisOCR2, Unlimited-OCR added); the changelog also notes a v1.7 update on 2026-04-30 but the table is still labelled v1.6_full. 35 rows: 33 models (specialised OCR VLMs and general VLMs) plus 2 pipeline tools moved to omnidocbench-tools. Maintained and broad on open OCR models, but its API/LLM entries are older versions (Gemini 3 Pro/Flash, GPT-5.2, Kimi K2.5, Mistral OCR 2503) and are not tracked. Re-checked 2026-10-19: last repository commit 2026-09-11 (leaderboard update adding TeleOCR, OvisOCR2, Unlimited-OCR); still no current Claude/GPT/Gemini rows.
ParseBench (models) by LlamaIndex (automated tests, Overall: mean of five rule-based dimension scores (tables, charts, content faithfulness, semantic formatting, visual grounding) on ~2,000 human-verified enterprise pages) · snapshot 9 Oct 2026
Model rows of leaderboard.csv in github.com/run-llama/ParseBench (full sortable board at parsebench.ai, also on Kaggle): 66 proprietary VLM rows, 47 open-weight VLM rows and 2 Mistral OCR 4 rows; several rows are effort/thinking settings of one model. Maintained and current: last leaderboard commit 2026-10-09 added LightOnOCR-3; frontier versions (Claude Opus 5.5, Fable 5.1, GPT-6.1 Sol, Gemini 3.8 Flash) are listed. Published by LlamaIndex, whose LlamaParse leads the tools view. Untracked older versions (GPT-5.x, Gemini 3.5-3.7 Flash, DeepSeek V4.1 Flash pre-1017, Mistral OCR 4) are not recorded.
MDPBench (multilingual document parsing) by MDPBench authors (Yuliang-Liu/MultimodalOCR) (automated tests, Overall: page-level mean of text, formula and table parsing scores on the public split (3,400 images, 17 languages, digital and photographed pages)) · snapshot 4 Oct 2026
34 model rows (9 general VLMs, 25 specialised OCR VLMs) of leaderboard.json in the Hugging Face space (updated 2026-10-04; additions in Aug-Oct 2026 incl. Kimi K3, U2-OCR, PepperOCR-VL). A held-out private split is also reported per row; the public overall is recorded. Maintained by the group behind MonkeyOCR and OCRBench, so MonkeyOCRv2 rows come from the publisher. Current versions tracked: Kimi K3, MonkeyOCRv2, dots.mocr, Chandra OCR 2, PaddleOCR-VL-1.6, HunyuanOCR-1.5, olmOCR 2, GLM-OCR, Nanonets-OCR2-3B, Falcon-OCR, Unlimited-OCR. Older versions (Gemini 3 Pro preview, Claude Sonnet 4.6, GPT-5.2, Mistral OCR 4.0) and the ambiguous 'MinerU-2.5-Pro-1.2B' row are not recorded.
OCR Arena by Extend (extend.ai) (blind human votes, Elo rating from blind pairwise human votes on OCR outputs of user-uploaded documents) · snapshot 19 Oct 2026
Rows read from the site's JSON endpoint (https://www.ocrarena.ai/api/leaderboard; its 'lastUpdated' equals the fetch time). The model set is frozen: Wayback snapshots of the endpoint from 2026-06-10 and 2026-07-21 list the same 24 entries (newest models: Gemini 3 Flash, Claude Opus 4.6, GPT-5.2, GLM-OCR), while votes continue slowly (per-entry battle counts sum to 33,073 vs 31,795 on 2026-06-10). Ratings are online Elo (K=20) and can move tens of points on a few dozen votes. No current frontier version is listed, so only OCR-specialist entries whose version is current are tracked (GLM-OCR, olmOCR 2, Nanonets-OCR2-3B). Battles per entry: 344 to 2,333.
OmniDocBench v1.6 (end-to-end, v1.6_full): pipeline tools by OpenDataLab (Shanghai AI Laboratory) (automated tests, Overall page-parsing score: ((1 − text edit distance) × 100 + table TEDS + formula CDM) / 3, ranks tools) · snapshot 11 Sep 2026
Rows typed 'Pipeline Tools' in the same table: MinerU-Pipeline (MinerU 3.4.0) and Marker (1.8.2). Other pipelines in the README's model-information table (Docling, PP-StructureV3, Unstructured, Mathpix) have no row on the v1.6 leaderboard. Re-checked 2026-10-19: unchanged.