Data Labeling Platform for ML

Top-quality data

Collect and annotate training data that meets and exceeds industry quality standards thanks to multiple quality control methods and mechanisms available in Yandex.Toloka.

Scalable projects

Have any amounts of image, text, speech, audio or video data collected and labeled for you by millions of skilled Yandex.Toloka users across the globe.

Cost-efficiency

Save time and money with this purpose-built platform for handling large-scale data collection and annotation projects, on demand 24/7, at your own price and within your timeframe.

Free, powerful API

Build scalable and fully automated human-in-the-loop machine learning pipelines with a powerful open API.

Yandex.Toloka Platform

Designed by engineers for engineers, Yandex.Toloka lets you integrate an on-demand workforce directly into your processes. Our cloud-based crowdsourcing platform is a fast and efficient way to collect and label large data sources for machine learning and other business purposes.

Start with raw data (text, images, URLs, video, audio, or any other type)

Configure the project and quality control, train and screen users

Get results with annotated data

Use Cases

Yandex.Toloka helps to improve models of any kind, including those in text & natural language processing, computer vision, chat bots and voice assistants, search and information retrieval, as well as offering solutions for business challenges and projects on any scale.

Image &Video Data

Text Data

Audio Data

Data Enrichment

Spatial Crowdsorcing

Object Recognition & Detection

Train your computer vision model by labeling image elements with bounding boxes, polygons or key points, as well as utilizing image segmentation and tagging based on your own ontology.
Price for 1000 tasks: $15.
Turnaround time: 3 hours.*

Image & Video
Classification

Collect a library of annotated images or videos — grade image quality, classify by type, identify objects or content, or receive any other judgments you need.
Price for 1000 tasks: $4.5.
Turnaround time: 1 hour.*

Image
Transcription

Get the text in PDF files annotated and transcribed to train your algorithms to better identify specific parts of documents, or validate and fine-tune the output of your own OCR models.
Price for 1000 tasks: $15.
Turnaround time: 3 hours.*

Side-by-Side
Comparison

Perform side-by-side comparisons to decide which option works best and check that your images match descriptions, or get any other judgments you need to verify or clean up your data.
Price for 100 image pairs starts at $1.5.
Turnaround time: 1 hour.*

Image & Video Moderation

Make sure inappropriate content isn’t uploaded to your site. Set rules for all types of images and videos with instructions and details about what is acceptable and what is not.
Price for 1000 tasks: $4.5.
Turnaround time: 1 hour.*

Image & Video
Collection

Create collections of images and videos based on specific topics or motifs, lighting, angles or environments. Ask crowd performers to record video snippets according to your specifications and instructions.
Price for 100 tasks: $6.
Turnaround time: 5 hours.*

Search
relevance

Use Yandex.Toloka to evaluate the performance of your search engine, find out which ranking model works best, and improve the search algorithm.
Price for 1000 tasks: $18.
Turnaround time: 4 hours.*

Text
Classification

Categorize any kind of text into predefined categories or ontologies to get the training data you need to optimize your NLP model.
Price for 1000 tasks: $18.
Turnaround time: 2 hours.*

Sentiment
Analysis

Categorize any kind of text by sentiment and reasons behind the sentiment for any purpose, from understanding customer reviews to spam filtering.
Price for 1000 tasks: $4.5.
Turnaround time: 1 hour.*

Intent
Classification

Train your chatbot, voice assistant, or any other conversational agent to better understand your user’s intent.
Price for 100 tasks: $6.
Turnaround time: 1 hour.*

Dialog
Collection

Power your conversational agent with text utterances collected by performers based on instructions or scenarios that you provide.
Price for 100 tasks: $12.
Turnaround time: 4 hours.*

Named Entity Recognition

Create your own custom ontology to identify parts of speech, classify proper nouns, or label any other entities for your project or named entity recognition (NER) model.
Price for 1000 tasks: $18.
Turnaround time: 1 hour.*

Audio Data Collection

Create or fine-tune a voice interface using speech samples recorded by Yandex.Toloka performers according to your instructions, with fast turnaround and low prices. Enhance TTS (Text-to-Speech) and speech synthesis technologies with high-quality audio data.
Price for 1000 tasks: $18.
Turnaround time: 3 hours.*

Audio Transcription

Convert audio to text with Yandex.Toloka transcribers or ask performers to check transcriptions for correctness and use resulting data to improve speech recognition models. Transcribe any number of audio files quickly and accurately.
Price for 1000 tasks: $7.
Turnaround time: 4 hours.*

Audio Classification

Use Yandex.Toloka to detect emotion, categorize topics, or identify events in audio samples or conversations to improve your model.
Price for 1000 tasks: $7.5.
Turnaround time: 2 hours.*

Business data

Collect information like website URLs or business hours, or categorize businesses by type, location, or size. Use Yandex.Toloka to enrich business data and gather the information you need.
Price for 1000 tasks: $7.2.
Turnaround time: 4 hours.*

Surveys

Conduct surveys of thousands of Yandex.Toloka performers with a variety of backgrounds using various types of survey formats.
Price for 100 tasks: $6.
Turnaround time: 1 hour.*

Content Generation

Generate any type or any amount of content, such as product descriptions, recipes or user manuals, for any kind of project, on-demand 24/7.
Price for 100 tasks: $6.
Turnaround time: 1 hour.*

Offline data collection

Gather information about businesses like name, phone number, menu, and opening hours. Digitize offline data for your projects and business goals.
Price per task starts at $0.05.
Turnaround time and cost depend on the location.*

Price monitoring

Get information about actual in-store prices on your products and competitors' discounts at specific retail outlets and adjust your pricing to match.
Price per task starts at $0.05.
Turnaround time and cost depend on the location.*

Merchandising

Check where your products are placed and the allocated share of shelf space. Compare this data with information about competitors' products.
Price per task starts at $0.10.
Turnaround time and cost depend on the location.*

Secret buyer

Get insights into staff performance, quality of service and demand for your products.
Price per task starts at $1.
Turnaround time and cost depend on the location.*

Foot traffic

Use Yandex.Toloka to learn about foot traffic and road traffic for planning ad campaigns and analyzing strategic retail points.
Price per task starts at $0.50.
Turnaround time and cost depend on the location.*

Ad monitoring

Collect data for analyzing and monitoring outdoor advertising. Track promo campaigns.
Price per task starts at $0.20.
Turnaround time and cost depend on the location.*

Approximate cost. Includes 20% Yandex.Toloka commission. Not a public offer. Price and turnaround time for tasks are set by the requester and depend on the type of task, input data, and other factors.

Success stories

Alexey Hahunov, founder and CTO of Dbrain, IT company and Y Combinator alumnus:

"Yandex.Toloka helped us tackle the challenge of recognizing handwritten text in documents. We send anonymous text fragments to Yandex.Toloka, where qualified performers rewrite the text and send the results back to the Dbrain system for further AI training. This meant we were able to resolve even the most difficult recognition cases for our customers."

Konstantin Simonchik, science director and co-founder of ID R&D:

"Using Yandex.Toloka, we collected the world's largest database of 200,000 unique photos and videos to secure biometric systems against hackers and train neural networks to distinguish real faces from fakes. We chose Yandex.Toloka because of the user-friendly interface, active performers and optimal quality for the price."

Dmitry Akimov, VisionLabs data engineer:

"Yandex.Toloka is an excellent tool for data processing. We use it to collect and annotate thousands of images and videos every month in order to improve the performance of our algorithms. We chose Yandex.Toloka because of the fast turnaround time and the active participation of performers."

Mikhail Burtsev, head of the DeepPavlov.ai laboratory for neural networks and deep learning at Moscow Institute of Physics and Technology:

"Modern machine learning technologies are not possible without a large volume of training examples. A computer can't master a natural language unless a person „explains“ how to use it correctly. Yandex.Toloka allows the conversational AI systems that we develop at MIPT to communicate with a large group of people and grow smarter."

Angelina Galkevich, Analyst
Tinkoff:

"We use Yandex.Toloka to annotate data for speech synthesis and recognition. Yandex.Toloka performers have already processed tens of thousands of tasks with audio files for us, and we have successfully used the results to train our models. We like the price, the fast turnaround, and the super fast feedback. It allows us to quickly test different markup methods and hypotheses, and then choose the optimal approach."

Anton Slesarev, head of Technologies in the self-driving cars division at Yandex:

"Toloka is my favorite Yandex service and the first place we go to prepare data for AI. To train the neural network, we need tens of thousands of annotated images. You can buy them for $4 each, or you can have images marked up in Yandex.Toloka 10 times cheaper and get datasets that are ready to use for training algorithms. We use these datasets for developing self-driving vehicle technology."

Yandex.Maps Team:

"Our service tracks outdated information on Yandex.Maps. We update business hours, add new businesses, and remove any that have closed. Yandex.Toloka users perform our tasks in their neighborhoods, like taking photos of business signs with operating hours. Every month we distribute 15 million tasks like this to 50,000 performers — this is the only way we can keep the huge Yandex.Maps database up to date."