Claude Sonnet 5 vs Unstructured
A raw model that fills in fields against a parsing layer that doesn't. Claude called directly came second in our benchmark. Unstructured, with the same field-filling step applied to its parse, came last. That's a category difference more than a quality one. For extraction, Claude. For a cheap, well-certified parsing layer feeding search, Unstructured is doing a different job that our test doesn't measure.
Claude here takes a file and a field list and returns the fields. Unstructured splits a document into elements for a retrieval system to index. To rank it we ran the same field-filling step over its output that every system got. The twenty-four-point gap between them is mostly the gap between a system that keeps a document's structure and one that hands it off.
On clean English text the two were close. Text files, spreadsheets, and simple invoices came through Unstructured nearly intact. Everywhere else Claude was far ahead, and on long documents it was second only to DocuPipe. Long registers lost their row structure in Unstructured's parse and had nothing to rebuild from. Chinese and Japanese scans came back nearly empty. A one-page table every other system read perfectly came back with nothing. Unstructured did hold up better than its rank suggests on Hebrew and Arabic text direction, and it was slightly better than Claude on photographed pages.
Resilience is a category problem for Unstructured in this test. It can't finish the job alone, and on long, tabular, or non-Latin documents the structure the next step needs is gone. Claude has nothing to tune and returned nothing empty.
On price both are cheap. Claude is a few cents a page on token billing. Unstructured is a cent and a half a page after 10,000 free pages, no card, no minimum, but that buys parsing only. A model step, possibly Claude itself, sits on top and costs extra.
On product Unstructured is the more complete for its purpose: a hosted platform that builds workflows from source connectors through partitioning, chunking, and enrichment to destinations, a Python SDK, REST API, MCP server, and a widely used open-source library, with zero data retention stated on every plan and SOC 2 Type 2, HIPAA, and GDPR on the pricing page. Claude offers a first-rate general API with native PDF input, Files API, batch, and structured outputs, and no document product, retention controls, or pipeline.
For structured extraction, Claude, clearly. For ingesting ordinary English documents into a search index with clean compliance terms, Unstructured, with a model like Claude behind it for any fields you need.
Structured extraction from documents of any kind, especially long, tabular, scanned, or non-Latin ones.
Search and retrieval pipelines over ordinary English documents where you want cheap parsing, connectors, and clean data terms.
How it scores, area by area
The accuracy score comes from the benchmark. Every other score is ours, with the reason written next to it. We use the same weights on every review and comparison.
| Area | Claude Sonnet 5 | Unstructured |
|---|---|---|
| Accuracy35% of total | 9.2Ranked 2nd of 8 systems on the benchmark. | 6.8Ranked 8th of 8 systems on the benchmark. |
| Resilience15% of total | 7.0No settings to tune and nothing returned empty, but a few ordinary forms came back badly misread with no warning, and there is no layout stage to catch them. | 4.0Needs your own extraction step to produce fields at all, and on long or tabular documents the structure is lost before that step runs. |
| Price15% of total | 9.0$2 and $10 per million tokens; a few cents per page at most. | 8.0$0.015 per page for parsing, but you add your own model cost to get structured fields. |
| Getting started10% of total | 3.0No document product. You build upload, review and results tooling yourself. | 6.0Hosted platform and open-source library, 10,000 free pages, no card; not an extraction product. |
| API and integration10% of total | 5.0Excellent general SDKs, native PDF input, a Files API, batch mode and structured outputs; no document jobs, webhooks or pipeline. | 6.0Python SDK, MCP server and a full workflow API with connectors, but built around partitioning rather than schema extraction, and webhooks are not marketed. |
| Data control7.5% of total | 5.0Platform policy only; zero retention for eligible organisations. Nothing document-specific. | 8.0Zero data retention stated on every plan. |
| Compliance7.5% of total | 6.0Platform certifications apply to the API; your pipeline is your responsibility. | 8.0SOC 2 Type 2, HIPAA and GDPR; dedicated instance and VPC on Business. |
| Overallweighted | 7.2weighted score out of 10 | 6.6weighted score out of 10 |
How each area is scored
- Accuracy (35%). The benchmark result, on a 0 to 10 scale.
- Resilience (15%). Does it just work? Fewer modes to choose, no tuning to reach the ranked result, no runs that fail outright, and no documents that come back quietly empty. Informed by the benchmark, judged by us.
- Price (15%). Published cost per page for parse plus extraction, cheapest scoring highest. Unpublished pricing is marked down.
- Getting started (10%). How far you get without a sales call: a dashboard, a free tier with no card, help authoring schemas, a test environment.
- API and integration (10%). Webhooks and which tier they are on, SDKs, no-code connectors, whether the API is built around document jobs at all.
- Data control (7.5%). Deletion controls, scheduled retention, zero-retention availability and on which tier, choice of region.
- Compliance (7.5%). Certifications stated on public pages: SOC 2, ISO 27001, HIPAA with a BAA, GDPR, private deployment options.
By document type
Average accuracy on each group of benchmark documents, both systems on the same files. Every document and score is on the evidence page.
| Document type | Claude Sonnet 5 | Unstructured |
|---|---|---|
| Statements, filings, reports2 to 5 pages · 21 documents | 92.3% Good | 84.9% Good |
| Short forms, invoices, receiptsone page · 30 documents | 91.7% Good | 80.2% Good |
| Long registers, catalogs, directories6 pages and up · 21 documents | 91.3% Good | 32.5% Weak |
| Non-English documents12 languages · 25 documents | 85.2% Good | 62.2% Mixed |
| Scans and photosimages, not PDFs · 7 documents | 80.1% Good | 86.5% Good |
Price, product, and policies
Full reviews: Claude Sonnet 5, Unstructured.
| Claude Sonnet 5 | Unstructured | |
|---|---|---|
| Pricing model | Per token through the Anthropic API. No document-specific product or plan. | Per page, parse only. You supply the extraction step yourself. |
| Entry price | $2 per million input tokens, $10 per million output tokens. | 10,000 free pages, no card. Then $0.015 per page pay-as-you-go. Business plan on request. |
| Cost per page | Depends on page size and how much you ask it to return; typically a fraction of a cent to a few cents per page. | $0.015 per page for parsing. Add the cost of whatever model you run on the output to get structured fields. |
| Free tier | No ongoing free tier for API use. | Yes. 10,000 pages, no credit card. |
| Pay as you go | Yes. Pure usage billing. | Yes. |
| Dashboard | No document product. A console workbench for prompts; upload, review and results tooling is yours to build. | Hosted platform for building workflows from source connectors (files, storage, databases) through partitioning and chunking to destination connectors, plus a widely used open-source library. |
| Webhooks | None for document jobs. You run the pipeline. | Workflow and job notifications through the API; not marketed as webhooks on the pricing page. |
| API and SDKs | Official SDKs in Python, TypeScript, Java, Go, Ruby, C# and PHP. PDF input up to 600 pages per request, a Files API, batch processing at reduced cost, structured outputs and citations. | Python SDK, REST endpoints, an MCP server, and workflow and job management APIs. |
| Data retention | Governed by Anthropic's API data policies; zero-retention arrangements exist for eligible organizations. Nothing document-specific. | Zero data retention stated across all plans. |
| Compliance | Anthropic's platform certifications apply to the API; your pipeline around it is your responsibility. | SOC 2 Type 2, HIPAA, GDPR. Dedicated instance and VPC options on Business. |
We checked these against the vendors' own pricing and documentation pages on 2026-09-11. Sources: Claude Sonnet 5: platform.claude.com; Unstructured: unstructured.io, docs.unstructured.io. Prices change, so confirm before you budget.
Frequently asked
Is Claude better than Unstructured for document extraction?
Yes, 91.73% to 67.67% in our benchmark. Unstructured is a parsing layer rather than an extraction product, and with the same field-filling step applied it held up only on clean English text.
Can I use Unstructured and Claude together?
That's a common pattern. Unstructured parses and chunks, and a model fills the fields. Our benchmark scored Unstructured's parse with a fixed extraction step applied, which is what that pattern measures.
Scores are macro-average field accuracy from the DocuBench 0.2.0, 72-document run. Every document, answer key, and raw extraction is open in the explorer; the scoring rules are on the methodology page, and the replication kit re-runs the scorer. A document a system could not process scores zero rather than being dropped.