You.com Extract API brings costly, slower PDF parsing
You.com’s Extract API parses PDFs with chart-aware accuracy, but standard mode costs $13.90 per 1,000 typical files and has an 11.0-second median latency.

Why it matters
- You.com launched Extract on November 17, 2026.
- The default PDF mode scored at least 80% higher than Exa Contents and Parallel Extract in You.com's five-dimension test.
- Extract standard scored 74.9 for tables and 76.6 for charts, compared with Exa's 22.8 and 0.9.
- Median latency was 11.0 seconds for standard and 2.6 seconds for lite.
- Lite costs $0.90 per 1,000 URLs; standard costs $13.90 per 1,000 typical 26-page PDFs.
By Rebecca Stone
You.com launched the Extract API on November 17, 2026, with a default PDF parser that the company says scored at least 80% higher than Exa Contents and Parallel Extract in a five-dimension evaluation.
You.com says Extract gives developers one endpoint to fetch a web page or PDF and return its contents as Markdown or HTML. The launch matters because agent builders often need an entire filing, contract or research paper after a search rather than a snippet. Developers can also call Extract directly when they already know the URL.
The launch puts a practical cost-and-latency trade-off in front of agent builders. Standard mode targets difficult documents; lite prioritizes speed and a flat rate for text already embedded in a PDF.
What does Extract retrieve?
Requests require an API key. You.com says responses identify whether the material came from a cache or a live fetch. Extract is designed for agents that need complete documents after search, but direct requests remain useful when a pipeline already holds the source URL.
The API reference defines three retrieval settings:
- The default
blendsetting uses cached content when available and crawls the page live when no cached copy exists. fetchalways crawls the live page.cachereturns stored content only. If no stored copy exists, the response omits the content field.
These controls separate freshness from availability. blend favors existing cache copies but falls back to a live crawl. fetch forces fresh content, while cache avoids a network request at the risk of returning no content field.
A request has a default end-to-end timeout of 30 seconds. You.com allows developers to configure that limit from 2 to 100 seconds. Once the limit expires, the endpoint returns a 504 error.
How do retrieval and PDF parsing differ?
PDF processing runs through a separate parse_effort control. The default standard setting sends complex pages, including charts and images, through an advanced model. The lite setting extracts only content embedded in each page.
You.com says standard mode is intended for tables, charts, forms and scanned pages. Its lite mode suits digitally created, text-based documents such as contracts. The You.com API reference states: “Neither setting changes web-page output or pricing.”
The controls address separate stages. Retrieval chooses whether a request uses cached content or crawls live; parse_effort determines PDF processing depth. A pipeline can therefore make both choices without altering the web-page output or the pricing basis.
What did the benchmark actually test?
You.com tested both Extract modes, Exa Contents and Parallel Extract across 3,782 pages in 12 ParseBench-2 sets. Each service received the same URL. ParseBench's grader scored the outputs, and any document a service failed to return received a zero.
You.com reported its largest advantages in tables and charts:
- Extract
standardscored 74.9 for tables and 76.6 for charts. - Exa Contents scored 22.8 for tables and 0.9 for charts.
- Parallel Extract scored 0.1 for tables and zero for charts.
Speed moved in the other direction. You.com reported the following median latencies:
- Extract
standard: 11.0 seconds. - Extract
lite: 2.6 seconds. - Exa Contents: 1.3 seconds.
- Parallel Extract: 1.8 seconds.
Parallel had a minimum cache age of 600 seconds, You.com said, so some of its responses may have been cached. That condition means the latency figure does not necessarily represent a live crawl for every request.
The zero rule prevents a non-return from being excluded from the comparison. It does not, however, turn the scoped evaluation into a complete ParseBench-2 leaderboard result. You.com disclosed the table and chart scores for standard mode, but the reported comparison did not provide corresponding table and chart figures for lite.
You.com's headline comparison describes its scoped test, not the full ParseBench-2 overall ranking. The company excluded the coordinate-based layout evaluation because Extract returns parsed content rather than page-element coordinates.
LlamaIndex's public leaderboard covers 56 methods, 4,317 pages and 13 datasets across six dimensions. Its overall score is the plain mean of those dimensions. You.com's five-dimension comparison therefore does not place Extract against every method in that broader ranking.
How much does Extract cost?
Lite costs a flat $0.90 per 1,000 URLs. Standard uses the same URL rate and adds $0.50 per 1,000 pages when the fetched file is identified as a PDF.
You.com based its comparison on a 26-page median across the billions of PDFs it has processed. At that document length, the company puts the price of 1,000 typical documents at $13.90 on standard. The same 1,000 URLs cost $0.90 on lite.
For comparison, You.com lists Exa and Parallel at $1.00 per 1,000 URLs regardless of document length. On the company's figures, standard mode carries a substantial premium for multipage PDFs, while lite undercuts both competitors for text-based files.
At 26 pages, the standard charge for a 1,000-file batch comes from the added PDF page rate. Lite keeps the same $0.90 cost as files grow in length. You.com's figures make standard most relevant when tables, charts or scans justify the wait, and lite more attractive for text-first retrieval.
You.com does not bill requests that fail on its side. Its error responses distinguish blocked URLs, failed requests and pages without content, allowing developers to decide whether a pipeline should retry.
Zero data retention is available for sensitive documents at no additional charge. Customers must contact You.com to enable the option.
For web pages, developers can use Extract through You.com's official MCP server for free and without an API key or authentication. Direct API sign-ups include $10 in credit, or $100 for customers signing up with a work email. You.com estimates the $100 balance covers about 7,000 typical PDFs processed in standard mode.
What changes for Contents API users?
You.com's transition guidance says: “The older Contents API remains supported for existing users but is no longer in active development.” Existing integrations can continue using it, but the company has reserved deeper PDF parsing and parse_effort for Extract.
The split leaves existing Contents users on a supported but frozen path. New document-depth controls belong to Extract, according to You.com's API transition.
For agent builders, the practical test is whether each document justifies standard mode's 11.0-second median latency and its $13.90 price per 1,000 typical 26-page PDFs, or fits lite's 2.6-second path at $0.90. With Contents feature development frozen, Extract becomes the control point for document-heavy retrieval.
Source: unite.ai
More from Rebecca Stone
Show full bio
Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.
728 articles
Related articles
- Exa 3.0 Claims First Place on 8 Search Benchmarks — All Run by Exa
- Databricks Puts GPT-5.5 to Work on Enterprise Agent Workflows
- AWS Adds Single-Shot RAG and Retrieval-Only Mode to Bedrock
- AWS Opens Direct Web Search API at $7 per 1,000 Queries
- OpenAI's Decisions API Enters Public Beta With 10x Faster Typed Answers