DocsRouter
DocsRouter is an early-access API service for extracting information from images and documents. It provides a single interface for comparing OCR-capable models through OpenRouter-powered inference. Developers can choose a model directly or route a request by an objective such as speed or cost.
Send a document through one endpoint
Users can upload an image or document, or supply a reachable file URL. The published example sends a document URL to the OCR endpoint with a balanced routing strategy and JSON output. The service describes OpenAI-compatible requests, which may help teams keep a familiar request structure while comparing document models.
A unified response contains extracted text, page output, structured fields and usage metadata. That is useful when an application needs a consistent response format even as the selected model changes. Confirm the fields your application needs with representative documents before relying on that schema for production processing.
Select a model or routing objective
DocsRouter offers quality, speed, cost and balanced strategies, as well as explicit selection by OpenRouter model ID. These options let a developer compare the tradeoffs of a document-processing workload without changing the main endpoint for each model.
A routing label is an objective, not a guarantee of correct extraction. Assess the result on your actual page layouts and fields, including documents that are harder to read. The captured site does not publish accuracy benchmarks, document-size limits or latency guarantees.
Usage fees and document conversion
The pricing page lists underlying provider cost plus a 5% DocsRouter fee for OCR processing. Accounts use pay-as-you-go billing, with $2 in new-account credit. API responses report estimated usage costs, and the dashboard provides cost tracking.
Office-document conversion is priced separately: cloud sandbox compute carries a 30% markup with a minimum charge of one cent per conversion. The publisher says Mistral OCR can handle DOCX and PPTX natively without conversion fees. Check whether your chosen model requires conversion, because that changes the cost beyond OCR inference itself.