Published by NutrientUpdated October 16, 2026

LlamaParse vs Reducto vs Extend

Q: Which is best for RAG ingestion: LlamaParse, Reducto, or Extend?

For born-digital documents at volume, LlamaParse Cost-effective returns Markdown at about $3.75 per 1,000 pages (our arithmetic); its Fast tier (about $1.25) returns spatial text without Markdown. For scans and forms where review matters, Reducto and Extend both return field-level citations and confidence. Nutrient's text mode at 1 credit per page is the fourth option to price.

Q: Which is cheapest?

For parsing at list: LlamaParse Fast about $1.25 per 1,000 pages, Reducto $10, Nutrient understand $25.20 at the Starter rate or $10.80 at the Pro rate (our arithmetic), Extend about $25 on pay-as-you-go. For extraction: Reducto $20, Nutrient understand extract 15 credits per page, LlamaParse 8 to 60 credits per page including parsing, Extend about $62.50 per 1,000 on pay-as-you-go. Plan rates change these.

Q: Which supports on-prem or air-gapped deployment?

Reducto lists both on Enterprise. LlamaParse lists self-hosting and BYOC on Enterprise. Extend lists BYOC and hybrid, and no on-prem option. Nutrient runs the same engine three ways: the hosted API; self-hosted in the Python and Java Data Extraction SDKs, where OCR and ICR run on your infrastructure with no external API calls and the VLM stage can run locally through Ollama, LM Studio, or vLLM; and on a Mac in Extract Studio, where documents never leave the machine when the model runs locally.

Q: Which has the most accurate extraction?

On Datalab's OmniExtractBench (run by Datalab, September 16, 2026), Reducto deep_extract scores 93.47 and Extend extraction_performance 90.17. LlamaExtract agentic_plus scores 84.93 on the same table. Extend's own RealDoc-Bench ranks Extend first. Nutrient is not on Datalab's OmniExtractBench leaderboard and publishes no extraction accuracy figure of its own. Run your own sample.

Q: Which should an AI agent builder pick?

All four have a hosted MCP server. LlamaParse's hosted MCP exposes Classify and Split. Nutrient's hosted MCP lists classify but not split, and it defaults to agentic mode at 18 credits per page, which matters on the free plan. Pick by the per-page cost of the mode the agent will actually use.

Q: Why is Nutrient on this page?

Nutrient published it. Nutrient Data Extraction competes on the same jobs, and the page keeps the rows where it trails so the comparison is usable.

Verdict

For production document parsing there is no single best API; the job decides. Choose LlamaParse when input breadth (130-plus formats) and a text tier at about $1.25 per 1,000 pages matter most. Choose Reducto when you want per-page USD list prices and private deployment up to air-gapped. Choose Extend when you want a workflow platform with review and evaluation around the API.

Choose Nutrient Data Extraction when you want one credit pool across parse, extract, classify, and split, grounded extraction that labels every field (id_match, fuzzy_match, not_found), and a free plan that renews 5,000 credits a month. Nutrient was founded as PSPDFKit in 2011. 3,000-plus organizations, including 15 percent of the Global 500, rely on it, with over a billion document interactions a year. The same extraction engine runs hosted, self-hosted in the Python and Java SDKs with no external API calls for OCR and ICR, or on a Mac in Extract Studio.

Every benchmark number on this page was run by a document-API vendor (Datalab, LlamaIndex, Extend, Reducto, or Nutrient), several of them in the table. Read them as directional and test on your own documents.

At a glance

LlamaParse (LlamaCloud), Reducto, Extend, Nutrient Data Extraction API; facts fetched 2026-10-15 to 2026-10-16
Fact	LlamaParse (LlamaCloud)	Reducto	Extend	Nutrient Data Extraction API
Product type	hosted API^[1]	hosted API^[2]	hosted API^[3]	Hosted Data Extraction API, web Studio, and macOS Extract Studio^[4]
Output formats	Markdown, Plain text, JSON (per-page items), XLSX, HTML tables, Annotated PDF, Structured JSON to custom schema^[1]	Markdown, JSON, HTML tables^[5]	Markdown, JSON, HTML tables^[6]	Spatial JSON, Markdown, Schema-shaped JSON with citations, Ranked labels with independent 0-1 scores, Split page ranges and ZIP files^[7]
OCR for scans	Yes^[8]	Yes^[9]	Yes^[10]	Yes: structure, understand, and agentic; text has no OCR; 100+ languages^[11]
Handwriting	Yes^[12]	Yes^[13]	Yes^[10]	Yes: printed-style in understand; cursive/freeform needs agentic^[14]
Tables	Yes^[1]	Yes^[13]	Yes^[10]	Yes: spans and bounds; complex tables may need agentic mode^[14]
Layout and bounding boxes	Yes^[15]	Yes^[13]	Yes^[16]	Yes: element and optional word bounds; word bounds exclude XLSX^[14]
Per-field confidence	Yes^[17]	Yes^[18]	Yes^[19]	Yes: 0-1 relative signal, not a probability^[20]
Source citations and grounding	Yes^[17]	Yes^[18]	Yes^[21]	Yes: field boxes, match labels, confidence, and source boxes^[20]
Schema extraction	Yes^[22]	Yes^[23]	Yes^[24]	Yes: closed JSON Schema subset; 128 KB, 2,000 nodes, depth 5^[25]
Classification	Yes^[26]	Yes^[27]	Yes^[28]	Yes: zero-shot, 2+ labels; independent 0-1 scores; 1 credit/page^[29]
Document splitting	Yes^[30]	Yes^[31]	Yes^[32]	Yes: boundary detection for PDF/images; 3 credits/page^[33]
Free tier	Includes 10K credits (Free plan, $0/month)^[1]	$150 in free usage (Standard plan)^[34]	10,000 free credits, then pay-as-you-go (no contract, monthly minimum or sales call)^[10]	5,000 credits/month; no card; all parse modes, including Auto^[35]
Pricing model	Credits per page by tier; 1,000 credits = $1.25 (i.e. $0.00125/credit). Parse v2 tiers: Fast 1, Cost-effective 3, Agentic 10, Agentic Plus 45 credits/page. Add-ons: layout extraction +3, enriched forms +10 per form page, spreadsheet 1/sheet, audio 3/min.^[36]	USD per 1,000 pages, per endpoint; Growth/Enterprise custom with volume discounts^[37]	Credits per page by operation and mode (Performance vs Light); plan sets the per-credit rate^[38]	Credits/page by endpoint and mode; Auto routes each page and bills that mode's rate; subscription plus capped overage^[39]
List price examples	Free: $0/month (10K credits; 5 concurrent parse jobs; community support) Starter: $50/month (40K credits; pay-as-you-go up to $500/mo (400K credits); basic email support) Pro: $500/month (400K credits; pay-as-you-go up to $5,000/mo; 20 concurrent jobs; priority Slack Connect support; limited-time one-time bonus of 800K credits ($1,000) for business email domains) Enterprise: Custom (Volume discount, 5x higher rate limits, SSO, SaaS or hybrid/VPC deployment, dedicated account manager, 100 concurrent jobs, custom BAAs)^[1]	Standard (list), Parse r-1: $10 / 1,000 pages (r-1 parse; OCR data return / Parse citations +$2/1k; prompted blocks +$5/1k) Standard (list), Extract: $20 / 1,000 pages (parsing included; dense extractions (>100 fields/page) may incur per-field surcharge) Standard (list), Deep Extract: $40 / 1,000 pages (agentic extraction; not eligible for batch discount) Standard (list), Split / Deep Split: $20 / $40 per 1,000 pages (parsing included) Standard (list), Classify / Deep Classify: $7.50 / $15 per 1,000 pages (first 5 pages by default; Deep Classify is alpha pricing) Standard (list), Edit: $60 / 1,000 pages ($15 for fully prefilled pages) (standalone, cannot be combined with other products) Batch queue: -20% on Parse and Extract (async, 12-hour completion guarantee) Growth / Enterprise: Custom (contact sales) (volume discounts, ZDR, BAA, EU/AU residency (Growth); VPC/on-prem, SSO/SAML, SLA (Enterprise))^[34]	Pay As You Go: $0.0125 per credit (10,000 free credits, all APIs, ZDR, no BAA, limited rate limits) Scale: $500 / month (50,000 credits/month included, then $0.01 per credit; Slack support; HIPAA/BAA add-on) Enterprise: Custom (BYOC and Hybrid, SSO/SAML, custom MSA/DPA/SLA, custom models, dedicated support) Per-page (Performance, PAYG): Parse 2 credits ($0.025); Extract 3 ($0.0375); Split 2 ($0.025); Classify 1 ($0.0125); Edit filling 1 ($0.0125); Form detection 2 ($0.025) (Extract also triggers an automatic Parse: pricing FAQ example is a 10-page extract = 50 credits (~$0.63)) Per-page (Light, PAYG): Parse 0.5 ($0.00625); Extract 0.7 ($0.00875); Split 0.5; Classify 0.3 (Light is lower accuracy on poor scans, hard handwriting, large tables) Operator extraction: 5 credits/page (agent-guided extraction, with rare overages)^[38]	Per-page credits: parse 1-18; extract 7-24 (classify 1; split 3) Auto mode: bills each page at the mode it selects (recommended starting mode; routes clean pages to structure and scans or complex forms to understand) Free: $0/month (5,000 credits; PAYG after a $25 wallet load) Starter: $59/month; $49/month annual (25,000 credits; $0.0028 per extra credit) Pro: $500/month; $450/month annual (500,000 credits; $0.0012 per extra credit)^[35]
Deployment	SaaS (US and EU regions), VPC / hybrid cloud, single-tenant SaaS, BYOC (Azure, AWS, GCP), self-hosting^[40]	SaaS, VPC, on-prem, air-gapped^[2]	SaaS, VPC^[3]	Hosted API: Nutrient-managed cloud, Self-hosted: Data Extraction SDK for Python and Java, no network required; OCR and ICR run on your infrastructure with zero external API calls, Local VLM: Ollama, LM Studio, or vLLM on your own hardware, or your own OpenAI or Anthropic key, Desktop: Extract Studio for Apple silicon Macs (macOS Tahoe 26.4+); the engine runs on the Mac and, with a local model, documents never leave it; Nutrient counts pages for credits, .NET data extraction SDK also offered^[41]
SDKs	Python (llama-cloud, PyPI 2.18.0 at fetch), TypeScript (@llamaindex/llama-cloud), Go, Java, CLI, REST API, n8n node^[42]	Python, Node.js, Go, REST API, CLI^[43]	Python, TypeScript, Java, Go, CLI^[44]	REST API, Python SDK, Java SDK, .NET page referenced, curl, JavaScript, and raw HTTP samples^[45]
MCP or agent tools	Hosted LlamaParse Platform MCP at mcp.llamaindex.ai/mcp (Parse, Classify, Extract, Split, Index); docs-search MCP; agent skills/plugins for Claude Code and Codex^[46]	MCP server (local: uvx mcp-server-reducto; hosted: https://mcp.reducto.ai/mcp), CLI, llms.txt and agent guide^[47]	Hosted MCP server (https://mcp.extend.ai/mcp, OAuth), extend CLI with 'extend-cli' skill, agents.md^[48]	Hosted MCP, Data Extraction Studio, and agent skills^[49]
Maximum file size or pages	{"Batches": "10,000 files per batch (Pro/Enterprise)", "Classify": "512 MB", "Extract": "100 MB, 500 pages per job", "Parse": "512 MB, no page cap", "Split": "512 MB (PDF/DOCX/PPTX)"}^[50]	100 MB direct upload; up to 5 GB via presigned URL; no page limits on Standard plan^[51]	Not published^[10]	Direct upload: 150 MB, URL input: 50 MB, URL timeout: 30 seconds, Processing: 5 minutes; HTTP 408 if exceeded, Page limit: not published, Test-key limits: lower, unpublished; Support-enabled^[52]
Compliance	SOC 2 Type II, GDPR, HIPAA (BAA for Enterprise)^[40]	SOC 2 Type II (all tiers), HIPAA with signed BAA (Growth and Enterprise), Zero Data Retention (Growth and Enterprise)^[53]	SOC 2 Type II (stated as controls aligned with SOC 2 Type II requirements; report via Trust Center), HIPAA with BAA (add-on on Scale, included on Enterprise, not on PAYG), GDPR (eu1 region, Frankfurt)^[54]	SOC 2: Type 2, audited annually, BAA: no public statement, Transport: HTTPS/TLS only, No-retention plans: files discarded after processing, Retention plans: runs stored; may train/evaluate models, Payments: Paddle^[55]
Public benchmarks	ParseBench public leaderboard (15 Oct 2026 snapshot) LlamaParse Agentic Plus 90.2 (#1), LlamaParse Agentic 87.0 (#2), LlamaParse Cost Effective 80.6 (#5) of 149 entries (overall 0 to 100), parsebench.ai (LlamaIndex), 2026-10-15 (source) ParseBench (LlamaIndex, ~2,000 pages, 167K rules) LlamaParse Agentic 84.9% overall at ~1.2 cents/page; Cost Effective <0.4 cents/page; formatting 85.2% (Docling 1.0%) (overall of 5 dimensions), LlamaIndex (vendor-authored; self-benchmark), 2026-04-13 (source) OmniParseBench LlamaParse headline 90.4 (text 88.2, tables 92.6, layout 90.4) (mean of text/tables/layout family pass rates), Datalab (competitor-authored), 2026-10-08 (source) OmniExtractBench llamaextract agentic_plus accuracy 84.93, precision 86.57, recall 93.13, coverage 616/620 (accuracy), Datalab (competitor-authored), 2026-09-16 (source)^[56]	Parse r-1 vs Reducto's previous agentic pipeline 20% reduction in error rate (error rate (preview)), vendor (Reducto), 2026-09-01 (source) Extraction benchmark commissioned by Reducto, run by micro1 (225 documents) Reducto Deep Extract: 100% coverage, 99.6% precision, 99.6% recall, 99.3% leaf accuracy; Extend MAX: 93.3% coverage, 86.4% precision, 92.7% recall, 92.8% leaf accuracy (coverage / precision / recall / leaf accuracy), vendor-commissioned (Reducto; conducted by micro1), 2026-09-17 (source) RD-TableBench (1,000 complex table images) Not published (hierarchical Needleman-Wunsch table similarity), vendor (Reducto), 2024-11-04 (source) RealDoc-Bench, Reducto rows (run by Extend) layout adjusted F1 0.759; Document Q&A 91.1% (Reducto Agentic) (adjusted F1 / Q&A accuracy), competitor (Extend), 2026-05-26 (source) Unstructured parsing benchmark, Reducto Agentic row (run by Unstructured) (page shows '2026 (no day shown)'; date here is the fetch date) adjusted CCT 0.812; table cell content accuracy 0.708; element alignment 0.595 (see source), competitor (Unstructured), 2026-10-15 (source)^[60]	RealDoc-Bench (1,500 layout samples; 1,359 Q&A prompts, 581 documents) Extend Parse 2.0: Q&A accuracy 95.7%, layout adjusted F1 0.847; Reducto layout F1 0.759, Reducto Agentic Q&A 91.1% (adjusted F1 / Q&A accuracy), vendor (Extend), 2026-05-26 (source) LongArray-Extract (45 PDFs) 99.2% mean per-document accuracy; 45 of 45 completed (mean accuracy, failed runs scored zero), vendor (Extend), 2026-06-02 (source) PoliTax Split (page shows '2026 (no day captured)'; date here is the fetch date) best harness F1 72.48% (Claude Opus 4.6); '+28.4 pts' largest F1 lift on homepage (split F1), vendor (Extend), 2026-10-15 (source) Reducto-commissioned micro1 extraction benchmark, Extend MAX row 93.3% coverage, 86.4% precision, 92.7% recall, 92.8% leaf accuracy (see source), competitor-commissioned (Reducto), 2026-09-17 (source)^[65]	ParseBench public row 76.4; paired nightly: 82.8 (overall 0 to 100), LlamaIndex, 2026-10-15 (source) ParseBench nightly 82.8; paired public row: 76.4 (overall 0 to 100), Nutrient, 2026-10-15 (source) opendataloader-bench, Nutrient's own run understand 0.933; agentic 0.930; structure 0.892; text 0.889 (overall 0 to 1), Nutrient, 2026-10-16 (source) opendataloader.org leaderboard Nutrient row (free CLI) 0.885; opendataloader hybrid 0.907 (overall 0 to 1), opendataloader.org, 2026-09-23 (source)^[66]

Methodology

Facts were taken from each vendor's pricing and documentation pages on October 15, 2026, and every table cell links to its source. Per-page dollar figures derived from credit prices are marked as our arithmetic. Where a vendor's pages disagree with each other, the table says so. Nutrient wrote this page, and rows where Nutrient trails are kept.

Pricing per 1,000 pages

LlamaParse: Fast $1.25, Cost-effective $3.75, Agentic $12.50, Agentic Plus $56.25 (our arithmetic from $1.25 per 1,000 credits); Extract costs 8, 25, or 60 credits per page including parsing ($10, $31.25, or $75 per 1,000 pages; our arithmetic). Reducto: Parse r-1 $10, Extract $20, OCR data or citations $2 extra. Extend: about $25 for a Performance parse and about $62.50 for an extraction page on pay-as-you-go at $0.0125 per credit (our arithmetic); plan rates are lower. Nutrient: parse 1 to 18 credits per page (Auto routes each page and bills at the selected mode's rate), extract 7 to 24, classify 1, split 3; at Nutrient's pay-as-you-go rates ($0.0028 per credit on Starter, $0.0012 on the $500-a-month Pro plan) an understand parse page is $25.20 or $10.80 per 1,000 (our arithmetic).

Grounding and confidence

All four return per-field citations with bounding boxes on extraction. LlamaParse adds a 0 to 1 confidence when enabled (not in spreadsheet mode). Reducto labels citations high or low and adds granular confidence in Extract v3. Extend returns logprob and OCR confidence per field. Nutrient returns a relative confidence per field (not a probability), a match label, and not_found for fields it could not place.

Deployment and compliance

LlamaParse: SaaS US and EU, BYOC and self-hosting on Enterprise; SOC 2 Type II, GDPR, Enterprise BAA. Reducto: SaaS, VPC, on-prem, air-gapped; SOC 2 Type II; BAA and zero retention on Growth and Enterprise. Extend: SaaS, BYOC, hybrid; controls aligned with SOC 2 Type II; BAA add-on on Scale. Nutrient: SOC 2 Type 2 audited, HTTPS only, no BAA statement on the public pages. Nutrient runs the same engine three ways: the hosted API; self-hosted in the Python and Java Data Extraction SDKs, where OCR and ICR run on your infrastructure with no external API calls and the VLM stage can run locally through Ollama, LM Studio, or vLLM; and on a Mac in Extract Studio, where documents never leave the machine when the model runs locally.

Published benchmarks, with who ran them

Nutrient's published benchmark results, as recorded on 16 October 2026: on LlamaIndex's ParseBench the public leaderboard row for Nutrient agentic is 76.4. Nutrient's own nightly run of 15 October 2026 scores 82.8 on the same test and is not on parsebench.ai yet; in that nightly run, understand mode scores 64.0. LlamaParse Agentic Plus leads the public leaderboard at 90.2. Nutrient is not on Datalab's OmniParseBench or OmniExtractBench public leaderboards. In Nutrient's own run on the opendataloader-bench corpus the API scores 0.933 in understand mode and 0.930 in agentic; on the public opendataloader.org leaderboard Nutrient's row (the free CLI) scores 0.885, behind opendataloader hybrid at 0.907.

Datalab OmniParseBench (October 8, 2026, Datalab-run; selected rows): Gemini 3.8 Flash 92.9, Datalab accurate 92.7, Reducto 90.8, LlamaParse 90.4, Mistral OCR 89.5, Extend 83.3, Azure 83.0. Datalab OmniExtractBench (September 16, 2026): Datalab accurate 93.85, Reducto 93.47, LlamaExtract 84.9. LlamaIndex ParseBench (public leaderboard, 15 October 2026): LlamaParse Agentic Plus 90.2, LlamaParse Agentic 87.0, Nutrient agentic 76.4 on its public row and 82.8 in Nutrient's own nightly run the same day, where understand mode scores 64.0. Extend RealDoc-Bench (Extend-run): Extend Parse 2.0 Q&A 95.7 percent, Reducto Agentic 91.1 percent. Reducto's own: r-1 cut error rate 20 percent versus its previous pipeline. Nutrient's own run on the opendataloader-bench corpus: 0.933 overall, against a public opendataloader.org row of 0.885 for its free CLI.

An agent and MCP row

LlamaParse: hosted MCP at mcp.llamaindex.ai with Parse, Classify, Extract, Split, and Index, plus skills for Claude Code and Codex. Reducto: local and hosted MCP (mcp.reducto.ai), CLI, llms.txt. Extend: hosted MCP at mcp.extend.ai with OAuth, CLI skill. Nutrient: hosted MCP at mcp.nutrient.io covering extract, parse, classify, convert, OCR, sign, and redact; it defaults to agentic mode at 18 credits per page, so ask the agent for a lighter mode on simple PDFs.

Where Nutrient differs

The other products on this page are parsers or extraction APIs. Nutrient's Data Extraction API is the same kind of product, built by a company that has shipped document SDKs to enterprises since 2011. The facts below come from Nutrient's own public pages.

Founded as PSPDFKit in 2011, and the SDKs developers knew as PSPDFKit continue as Nutrient SDKs. Thousands of organizations in 80-plus countries, including Fortune 500s and governments, use them (about page).
3,000-plus organizations and 15 percent of Global 500 companies rely on Nutrient, with over a billion document interactions a year (nutrient.io).
One engine, four deployments: hosted API in Nutrient's cloud; the Python and Java Data Extraction SDKs, self-hosted with no network required and no external API calls for OCR and ICR, with a local VLM through Ollama, LM Studio, or vLLM (Python SDK); Extract Studio on Apple silicon, where documents never leave the Mac when the model runs locally (Extract Studio); and the wider platform in the browser with WebAssembly, on your own servers with Document Engine, or air-gapped (nutrient.io).
Extraction is one credit pool with parse, classify, and split, and every extracted field carries a bounding box and a match label, so a reviewer can check the source instead of trusting the model. SOC 2 Type 2 audited.

What to test before choosing

Run the same 20 documents through each API: five clean born-digital PDFs, five scans, five with tables that span pages, five with the fields your downstream system needs. Score reading order, table cells, and the fields you will actually consume.
Count the per-page cost at the mode or tier that passed your sample, not the cheapest tier on the pricing page. Every vendor here prices accuracy upward.
Check what the API returns when it is unsure: a confidence number, a bounding box, a not_found label, or nothing. That decides whether you can build a review queue.
Confirm retention and deployment in writing for the plan you will buy. Free and pay-as-you-go tiers often exclude BAAs, zero-retention, and private deployment.

Frequently asked questions

Which is best for RAG ingestion: LlamaParse, Reducto, or Extend?