Yandex Vision OCR
aistudio.yandex.ru
- Web
- API
What Yandex Vision OCR is
Yandex Vision OCR recognizes printed and handwritten text in images and PDFs, including pages with columns, tables, mathematical formulas, and document templates. Separate models cover general page text, multi-column text, handwriting, tables, Markdown text, and math formulas. Document models can recognize passports, driver licenses, vehicle registration certificates, and license plates and return template fields.
Requests use gRPC or REST APIs. Results are returned as JSON Lines and can include recognized text, language, page and text positions, image positions, and layout types such as titles, headers, footnotes, and lists. Supported files are JPEG, PNG, and PDF. The maximum API file size is 10 MB and maximum image size is 20 megapixels; asynchronous PDF requests can include up to 200 pages. Results remain on the server for three days. One language model is used per recognition operation, so multilingual text may need separate requests. Each successful image or PDF page is billed as one unit, without VAT: printed text and license plate recognition are listed as free, while the listed document, handwriting, and table recognition plans are $0.01, paid once. The service is cloud-based and has no free plan.
Who it suits
It suits developers and organizations that need OCR through an API for images or PDFs, including structured documents and tables. Teams processing multiple languages or large files should account for the per-operation language model and listed input limits.
Good to know, and what to watch
- Recognizes printed and handwritten text.
- Models cover tables, formulas, and document templates.
- Provides gRPC and REST API access.
- Asynchronous PDF requests support up to 200 pages.
- Successful printed text and license plate requests are listed as free.
- No free plan is listed.
- One language model is used per operation.
- API files are limited to 10 MB and images to 20 megapixels.
- Results are retained on the server for three days.
Plans
Every plan and priceLicense plate recognition
Printed text recognition
Document recognition (driving license)
Document recognition (passport)
Getting started
- Open the Yandex Vision OCR documentation website
- Use the gRPC or REST API interface
- Assign the ai.vision.user role for image analysis and related cloud access
- Send JPEG, PNG, or PDF files within the documented size limits
- Process the returned JSON Lines recognition data
Limits to check
Synchronous requests are limited to one per second and asynchronous requests to ten per second. Files are limited to 10 MB, images to 20 megapixels, asynchronous PDFs to 200 pages, and recognition results are retained for three days.
The Club's review
Yandex Vision OCR is a cloud API for turning printed or handwritten content in images and PDFs into structured recognition data. It suits developers, medium-sized businesses and larger organizations building document workflows in Yandex Cloud. The main attraction is layout-aware and document-specific output; the trade-offs are per-image billing, one language model per operation and strict file and throughput limits.
Overview
Results include recognized text plus page, block, line, word and image positions. Vision OCR groups words into lines, lines into blocks and blocks into pages, and can label elements such as titles, headers, footnotes and lists. That hierarchy is valuable when a downstream system must preserve reading order or reconstruct a form rather than store a text blob.
REST and gRPC interfaces return JSON Lines data, making the service suited to automated pipelines rather than a standalone office application. JPEG, PNG and PDF are supported. A request uses one language model, so documents containing multiple languages may need separate operations. Files are limited to 10 MB and images to 20 megapixels; asynchronous PDF requests can contain up to 200 pages. Results remain on the server for three days, so applications must collect them within that window.
Key features
Models for difficult layouts
Separate models handle general page text, multi-column pages, handwriting, tables, Markdown and mathematical formulas. This range makes Vision OCR a stronger fit than basic text-only OCR for mixed document collections, provided the workflow selects the appropriate model for each input.
Template and positional extraction
Document models recognize passports, driving licenses, vehicle registration certificates and license plates, returning fields from those templates. Positional metadata and layout types support indexing, validation and reconstruction. The service does not provide object detection, image classification, face recognition or custom models, so it is focused on OCR rather than a general computer-vision platform.
Pipeline controls
Successful image-recognition requests are chargeable units, with every image or page in a multi-page PDF counted separately. Failed requests caused by server errors or incorrect configuration are not charged. Quotas are one synchronous request per second and 10 asynchronous requests per second, which may require queueing for bursty workloads. Yandex documents an integration pattern using Object Storage, Cloud Functions and Serverless Containers, and the ai.vision.user role permits analysis and access to relevant cloud, folder and quota information.
Pricing
There is no free plan. Prices below are per unit, billed once and shown without VAT; each image or PDF page is a separate unit, and only successful requests incur a charge.
| Plan | Price | Best fit |
|---|---|---|
| Printed text recognition | Free | General printed pages |
| License plate recognition | Free | Plate template extraction |
| Document recognition (passport) | $0.01, paid once | Passport fields |
| Handwriting recognition | $0.01, paid once | Handwritten pages |
| Document recognition (vehicle registration certificate) | $0.01, paid once | Registration documents |
| Document recognition (driving license) | $0.01, paid once | Driving-license fields |
| Table recognition | $0.01, paid once | Tabular content |
The zero-priced printed-text and license-plate units still use per-request accounting and do not create a free tier for other models. The $0.01 document, handwriting and table units are inexpensive for occasional processing, but page-by-page charging can matter at high volume.
Platforms
Vision OCR is delivered as a cloud service through web-accessible Yandex Cloud tooling, with REST and gRPC APIs. JSON output, cloud deployment and documented proxy handling suit server-side integrations. Security information from Yandex Cloud describes certifications to ISO/IEC 27001, 27017, 27018 and 42001:2023, alongside stated personal-data safeguards and a shared-responsibility security model.
Who it's for
Choose it when a development team needs OCR embedded in an API workflow and values coordinates, layout labels, handwriting, tables, formulas or fixed document templates. It is also a practical match for organizations already operating Object Storage, Cloud Functions or Serverless Containers. It is less suitable for multilingual files processed in a single call, inputs beyond the size and page limits, workloads requiring sustained rates above the quotas, or teams seeking local deployment and custom models.
Pros and cons
Pros
- Rich structure: positional data and block types preserve document context for downstream processing.
- Broad recognition coverage: handwriting, tables, formulas, multi-column pages and named identity documents are covered by dedicated models.
- Predictable unit billing: each image or PDF page is counted, while failed requests are not charged.
- API flexibility: REST and gRPC access and JSON Lines output fit automated services.
Cons
- Single-language operations: multilingual documents may require multiple requests.
- Hard input constraints: 10 MB files, 20-megapixel images and 200-page asynchronous PDFs limit oversized archives.
- Throughput ceilings: one synchronous and 10 asynchronous requests per second can force queueing.
- Cloud-only scope: no custom models, object detection, classification or face recognition are included.
- Short retention: results are held for three days, requiring prompt retrieval.
Alternatives
For a broader computer-vision directory, see AI Image Recognition Software. For OCR-focused comparisons, see AI OCR Software.
- Sighthound AI is the better direction if you need Android, iOS, desktop, self-hosted and web options, plus a free plan and trial.
- Face++ suits buyers wanting a freemium API with free-key access and variable QPS limits.
- Google Cloud Vision API is an alternative freemium cloud API with pay-as-you-go image billing.
- SentiSight.ai is worth considering for an offline model license and self-hosted deployment.
- CompreFace is a free, open-source, Docker-based option with CPU and GPU support.
- InsightFace offers free open-source use for research and non-commercial projects.
- MediaPipe provides free on-device ML solutions across major platforms.
- Roboflow is a freemium platform with a free monthly credit tier for training and inference.
Verdict
Yandex Vision OCR is a strong choice for API-led document processing that needs coordinates, layout semantics and specialized models for handwriting, tables or standard identity documents. Choose it when Yandex Cloud integration and per-page accounting fit your pipeline. Look elsewhere if you need multilingual single-pass recognition, local or custom-model deployment, broader vision tasks or higher sustained throughput.
The verdict
Yandex Vision OCR covers varied page layouts and document types, with structured positional results for API workflows. The per-page billing, single-language operation, file limits, and three-day result retention shape how it may fit a processing pipeline.
Questions people ask
Does Yandex Vision OCR have a free plan?
No. It uses paid recognition plans, with printed text and license plate recognition listed as free and several other types at $0.01, paid once.
Which file formats are supported?
JPEG, PNG, and PDF files are documented as supported.
How is OCR accessed?
The service is available through gRPC and REST APIs.
What does each recognition result contain?
JSON Lines output can include text, recognition language, positions for pages, blocks, lines, words, and images, plus layout types.
Can it recognize passports and driving licenses?
Document-template models cover passports, driving licenses, vehicle registration certificates, and license plates, returning template fields.
What the maker publishes
- Access control
- The ai.vision.user role allows users to analyze images with Vision OCR and view relevant cloud, folder, and quota information. source
- API access
- The service is accessible through gRPC and REST APIs. source
- Cloud responsibility
- Yandex Cloud describes security as shared responsibility, with responsibility depending on the service model and security tools and policies in use. source
- Document extraction
- Document template models can recognize passports, driver licenses, vehicle registration certificates, and license plates and return template fields. source
- Failed requests
- Only successful analysis attempts are chargeable; server errors and incorrectly configured requests are not charged. source
- File formats
- Supported file formats are JPEG, PNG, and PDF. source
- Input limits
- The maximum OCR API file size is 10 MB and the maximum image size is 20 megapixels. source
- Input limits
- The maximum file size is 10 MB and the image size must not exceed 20 megapixels. source
- Integrations
- A Yandex Cloud tutorial shows an image and PDF workflow using Object Storage, Cloud Functions, and Serverless Containers with Vision OCR. source
- Intended customers
- Yandex Cloud says its platform services are for large companies, medium-sized businesses, and private developers. source
- Interfaces
- Vision OCR is accessible through gRPC and REST APIs. source
- Languages
- The service uses one language model per recognition operation, so text in multiple languages may require separate requests. source
- Layout
- The service groups words into lines, lines into blocks, and blocks into pages, and identifies block types such as headers, footnotes, titles, and lists. source
- Maker launch
- Yandex Cloud says it launched in 2018. source
- Output
- Recognition returns JSON Lines data with page, block, line, word, and image positions; words include text and recognition language. source
- PDF limit
- Asynchronous OCR API requests support PDFs of up to 200 pages. source
- Personal data
- Yandex Cloud says it implements personal data protection measures meeting first-level UZ-1 requirements under Russian Resolution No. 1119 and FSTEC Order No. 21. source
- Pricing basis
- Each successful image recognition request is charged as one pricing unit, with each image or page of a multi-page PDF charged separately. source
- Recognition models
- Models are available for general page text, multi-column text, handwriting, tables, Markdown text, and math formulas. source
- Recognition output
- Recognition results include text and page, block, line, and word positions, and can include image positions and layout types. source
- Request quotas
- The documented quotas are 1 synchronous request per second and 10 asynchronous requests per second. source
- Result retention
- Recognition results are retained on the server for three days. source
- Security compliance
- Yandex Cloud states that its information security management system is certified to ISO/IEC 27001, 27017, 27018, and 42001:2023. source
- Support
- Yandex Cloud documentation says users can request information about resource operations from cloud logs by contacting support. source
Checked against the maker's pages on 8 Oct 2026.