---
type: analysis
slug: guide-document-parsing-for-rag-llamaparse-unstructured-docling-mistral-ocr
title: "Document parsing for RAG: LlamaParse, Unstructured, Docling, Mistral OCR"
description: "For RAG document parsing (checked 2026-11-06): LlamaParse is a managed cloud API (130+ formats incl. PDF/scans/Office/slides/spreadsheets) with v2 tiers Fast/Cost-effective/Agentic/Agentic Plus billed in credits (10k free/mo on Pay-as-you-go); outputs markdown/text/JSON. Unstructured SaaS Transform"
published_at: 2026-11-16T13:00:00.000Z
topics: ["tools","developer_tools"]
source_urls: ["https://www.llamaindex.ai/llms.txt","https://docs.unstructured.io/open-source/introduction/overview","https://docling-project.github.io/docling/usage/mcp/","https://developers.llamaindex.ai/llamaparse/general/supported_document_types/","https://developers.llamaindex.ai/llamaparse/general/billing/","https://developers.llamaindex.ai/llamaparse/general/pricing/","https://unstructured.io/pricing","https://reducto.ai/pricing","https://azure.microsoft.com/en-us/pricing/details/document-intelligence/","https://developers.llamaindex.ai/llamaparse/parse/guides/tiers/","https://docs.unstructured.io/concepts/partitioning","https://docling-project.github.io/docling/usage/advanced_options/","https://docs.mistral.ai/studio/document-processing/annotations","https://docs.reducto.ai/reference/credit-usage","https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/overview?view=doc-intel-4.0.0","https://github.com/docling-project/docling/releases","https://docs.mistral.ai/models/ocr-4-1","https://docs.mistral.ai/admin/billing-usage/usage-limits"]
author: "Andrey Reshetnikov"
author_profile: https://newruntime.com/owner-profile.md
---
For RAG document parsing (checked 2026-11-06): LlamaParse is a managed cloud API (130+ formats incl. PDF/scans/Office/slides/spreadsheets) with v2 tiers Fast/Cost-effective/Agentic/Agentic Plus billed in credits (10k free/mo on Pay-as-you-go); outputs markdown/text/JSON. Unstructured SaaS Transform is flat USD $0.015/page after 10,000 free pages (Auto/Fast/High-Res/VLM strategies; markdown or elements JSON); OSS library exists for local prototyping. Docling (IBM/LF AI) is free local OSS (pip/CLI/MCP/docling-serve) for PDF/DOCX/PPTX/XLSX/images/scans with Markdown/HTML/JSON exports—no per-page cloud price on its docs. Mistral OCR API (mistral-ocr-4-1 / mistral-ocr-latest) returns per-page Markdown/JSON via /v1/ocr at USD $4/1,000 pages ($5 annotated); Free Studio mode has no API access. Alternatives: Reducto r-1 Parse USD $10/1k pages (r-1 mini $1/1k; $150 free credits); Azure Document Intelligence free 500 pages/mo then S0 Read USD $1.50/1k pages (then $0.60 after 1M), Prebuilt/Layout $10/1k, Custom extraction $30/1k (eastus retail).

*Field guide compiled 2026-11-06 from the official documentation, pricing pages and changelogs linked below; every line links to the page that states it.*

## What it does

- **LlamaParse accuracy claim (own docs).** LlamaIndex llms.txt (as of 2026-10-29): LlamaParse Agentic Plus ranks first on ParseBench (parsing); Extract Agentic Plus first on ExtractBench. Benchmarks are open source and built by the LlamaParse team. [source](https://www.llamaindex.ai/llms.txt)
- **Unstructured local vs API.** Open-source library for local prototyping (partition/chunk; no GPU support stated). Production: Unstructured API Transform/Pipelines (SaaS, dedicated, or in-VPC) with fine-tuned OCR/VLMs, enrichments, connectors. OSS lacks latest VLMs, fine-tuned OCR, by-page/by-similarity chunking, and compliance certifications listed for the API. [source](https://docs.unstructured.io/open-source/introduction/overview)
- **Docling local / API / MCP.** Runs locally (macOS/Linux/Windows; air-gapped via model prefetch). MCP server (uvx docling-mcp-server) converts locally by default or can call a remote docling-serve via DOCLINGSERVICEURL / DOCLINGCONVERSIONMODE=remote. No cloud per-page price on Docling project docs—cost is your compute. [source](https://docling-project.github.io/docling/usage/mcp/)
- **Supported formats detail — LlamaParse list.** Official common formats: documents pdf/docx/doc/pptx/ppt/rtf/txt/epub; images jpg/png/gif/bmp/tiff/webp/heic; spreadsheets xlsx/xls/csv/tsv/numbers/ods; audio mp3/mp4/wav/m4a/webm—plus many more (130+). [source](https://developers.llamaindex.ai/llamaparse/general/supported_document_types/)

## Modes

- **LlamaParse Fast / Cost-effective / Agentic / Agentic Plus (+ Cost Optimizer).** Required tier+version on every v2 parse. Fast=no AI; Cost Optimizer routes simple pages to costeffective under Agentic/Agentic Plus. [source](https://developers.llamaindex.ai/llamaparse/parse/guides/tiers/)
- **Unstructured Auto / Fast / High Res / VLM.** Auto recommended; routes PDF pages Fast vs VLM/High Res by content. Same SaaS page price on pricing page regardless of strategy. [source](https://docs.unstructured.io/concepts/partitioning)
- **Docling standard PDF pipeline / VLM pipeline / Native PDF / OCR engines.** PdfPipelineOptions (OCR EasyOCR/Tesseract/RapidOCR/…, TableFormer FAST/ACCURATE); CLI --pipeline vlm; NativePdfPipeline without layout/OCR models; local or remote serve. [source](https://docling-project.github.io/docling/usage/advanced_options/)
- **Mistral OCR basic + Annotations.** Basic OCR markdown/JSON; optional includeblocks, confidencescoresgranularity; bboxannotation and documentannotation for schema JSON (annotated page pricing). [source](https://docs.mistral.ai/studio/document-processing/annotations)
- **Reducto Parse r-1 vs r-1 mini.** r-1 highest accuracy/lowest latency list price; r-1 mini cost-focused. Extract/Split/Classify/Edit separate endpoints; parsing included in Extract/Split prices. [source](https://docs.reducto.ai/reference/credit-usage)
- **Azure Read / Layout / Prebuilt / Custom.** Read=text OCR; Layout=text+tables+structure; many prebuilt models (invoice, receipt, ID, tax, mortgage…); custom neural/template/classifier. Containers available; pricing page says container pricing same as cloud. [source](https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/overview?view=doc-intel-4.0.0)

## Pricing and limits

- **LlamaParse Pay as you go:** USD $0/mo subscription; credits USD $1.50 per 1,000 purchased. 10,000 free credits each month (unused monthly credits do not roll over). Concurrent Parse/Extract jobs: 5. Stored files: 50,000 / 50 GB (lower until first credit purchase). All products (Parse, Extract, Classify, Split, Index). [source](https://developers.llamaindex.ai/llamaparse/general/billing/)
- **LlamaParse Pro:** USD $500/mo; credits USD $1.25 per 1,000 purchased. 400,000 monthly credits; concurrent Parse/Extract 20; 100,000 files / 100 GB; Priority support. [source](https://developers.llamaindex.ai/llamaparse/general/billing/)
- **LlamaParse Parse v2 credits per page:** Fast 1 credit/page; Cost-effective 3; Agentic 10; Agentic Plus 25 (plus optional Layout extraction +3/page on pricing table; spreadsheet 1 credit/sheet; audio 3 credits/minute). Credits billed per page as listed; convert with plan credit price ($1.50/1,000 on Pay as you go, $1.25/1,000 on Pro). Re-parsing the same file within 48 hours costs 0 credits by default (cache). [source](https://developers.llamaindex.ai/llamaparse/general/pricing/)
- **Unstructured Free / Pay-As-You-Go:** USD $0 then USD $0.015 per page. 10,000 free pages to start (no card required on Free). After that $0.015/page; all features included on listed plans. Page = PDF page / PPTX slide / TIFF image; DOCX uses page metadata when present; else file size/100 KB. [source](https://unstructured.io/pricing)
- **Reducto Standard list (effective 2026-09-01):** Parse r-1 USD $10 / 1,000 pages; r-1 mini USD $1 / 1,000 pages; Extract $20; Deep Extract $40; Split $20; Deep Split $40; Classify $7.50; Edit $60 (or $15 prefilled) per 1,000 pages. Standard includes USD $150 free usage credits. Batch queue −20% on Parse (r-1/r-1 mini) and Extract; 12-hour completion guarantee. Concurrent pages r-1: 500 (Standard). [source](https://reducto.ai/pricing)
- **Azure Document Intelligence Free F0:** USD $0. 0–500 pages free per month (Free - Web/Container). Does not support premium features / Query Meter per pricing footnotes. [source](https://azure.microsoft.com/en-us/pricing/details/document-intelligence/)

## Recent changes

- 2026-11-05: Docling v2.148.0: Surya v2.5; PDF superscript/subscript from glyph size/baseline; iCalendar (.ics) backend. [source](https://github.com/docling-project/docling/releases)
- 2026-11-03: Docling v2.146.0: Unlimited-OCR 2 preset; LightOnOCR-3 presets; guarded Nemotron 3.5 multi-pass parsing. [source](https://github.com/docling-project/docling/releases)
- 2026-10-29: LlamaIndex states Agentic Plus #1 on ParseBench / ExtractBench as of this date (own llms.txt). [source](https://www.llamaindex.ai/llms.txt)
- 2026-09-01: Reducto per-product list pricing (r-1 / r-1 mini Parse etc.) goes into effect; see migration guide vs legacy credits. [source](https://docs.reducto.ai/reference/credit-usage)
- 2026-07-16: Mistral OCR 4.1 released: paragraph-level bboxes, structural block labels, block confidence; $4/$5 per 1k pages (standard/annotated). [source](https://docs.mistral.ai/models/ocr-4-1)
- 2026-03-30: Azure Document Intelligence v3.0 API announced retiring with end of support March 30, 2029; migrate to v4.0 (2024-11-30 GA). [source](https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/overview?view=doc-intel-4.0.0)

## Alternatives

- **Reducto.** Agentic document platform with Parse/Extract/Split/Classify/Edit APIs; 30+ file types; Studio; SaaS/VPC/on-prem. List Parse from $1–$10 per 1,000 pages (mini vs r-1) vs LlamaParse credit tiers / Unstructured $0.015/page / Mistral $4/1k OCR pages. [source](https://reducto.ai/pricing)
- **Azure Document Intelligence.** Microsoft cloud (+ containers) with Read/Layout/prebuilt/custom models and free 500 pages/mo; strong form/field extraction schemas. Per-1k-page S0 meters rather than credit packs; not Markdown-first like LlamaParse/Mistral OCR defaults. [source](https://azure.microsoft.com/en-us/pricing/details/document-intelligence/)

## Traps

- LlamaParse is commercial SaaS (not open source); Pay-as-you-go free credits are 10k/month and do not carry over. Jobs pause at zero credits unless auto-reload / Keep going is configured. Cache 48h (donotcache=True to skip). Starter plan no longer sold to new customers. [source](https://developers.llamaindex.ai/llamaparse/general/billing/)
- LlamaParse Fast runs no AI model—custom prompts, forms, and chart extraction do not apply. FAQ also notes ~45s average per page-block (varies). [source](https://developers.llamaindex.ai/llamaparse/parse/guides/tiers/)
- Unstructured Fast skips images and poorly handles handwriting/complex tables in PDFs; High Res often fails handwriting/multilingual; use VLM/Auto for those. Page billing rules differ by filetype (PDF/PPTX/TIFF = pages; else ~100 KB = 1 page). [source](https://docs.unstructured.io/concepts/partitioning)
- Mistral Free Studio/Vibe mode does not include API access; OCR requires billing activation. Document annotation path sends OCR markdown plus only the first eight extracted image bounding boxes to the vision model. [source](https://docs.mistral.ai/admin/billing-usage/usage-limits)
- Docling remote picture-description/API VLM and similar remote services require explicit enableremoteservices=True or OperationNotAllowed is raised—default is local-only user data. [source](https://docling-project.github.io/docling/usage/advanced_options/)
- Azure public pricing HTML often masks dollar amounts as $-; Free tier is explicitly 0–500 pages/month. Use Azure Retail Prices API or calculator for current S0 USD meters (eastus Read $1.50/$0.60, Pre-built $10, Custom $30 per 1k pages as of this fetch). [source](https://azure.microsoft.com/en-us/pricing/details/document-intelligence/)
- Reducto list prices on docs take effect 2026-09-01 (legacy credit chart differs). r-1 mini does not process spreadsheets. Dense Extract (100 fields/page) may incur per-field surcharge. [source](https://docs.reducto.ai/reference/credit-usage)

Author: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Tools](https://newruntime.com/topics/tools.md): Explore the tools topic hub.
- topic hub: [Developer Tools](https://newruntime.com/topics/developer-tools.md): Explore the developer_tools topic hub.
- collection: [Analysis](https://newruntime.com/analysis.json): Open the latest authorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)
