Extract
Neural Network
Extract by ScaleDown is an entity extraction model where fields are defined by your description, not a fixed list of types.
Specifications
- Input
- text, images, documents
- Output
- text
- Context size
- 1 000 000 tokens
- Max answer length
- 900 000 tokens
Prompt cost
(per 1M tokens)
How it works Extract?
Extract is a ScaleDown model designed for entity extraction: instead of a rigid set of types, you describe the required fields in the entities parameter, and the model returns their values from the provided text. This approach eliminates the need to adapt to every new schema—change the field description to get a different response structure. In BotHub, the model works from Russia without a VPN or foreign card, payment is in rubles based on usage, and tokens do not expire. Requests go through a unified OpenAI-compatible API, so Extract connects to your service by simply changing the model name, and you can switch to another of the 250+ neural networks without rewriting the integration. Data is transmitted with AES-GCM encryption and is not stored. It is useful when you need to parse incoming emails and requests into CRM fields, extract parties, amounts, and deadlines from contracts, turn forms and resumes into a table, or collect mentions of companies and products from reviews. For companies, contracts, invoices, and EDI are available, along with an admin panel with employee limits.Frequently asked questions about Extract
You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.
Extract accepts text, images, and documents as input, including native file processing, and returns a text result. A 1,000,000 token context window allows you to upload large reports, contracts, and scans to extract necessary information and collect structured summaries without manual labeling.
The output limit is 900,000 tokens per response, so the model easily provides voluminous results: full document transcripts, detailed summaries of hundreds of pages, or large data tables. You likely won't need to split the task into parts for length reasons.
A separate reasoning mode for Extract is not noted in our data—this doesn't mean it doesn't exist, we just don't confirm it. In practice, a simple trick helps: ask the model to break down the task step-by-step directly in the prompt. If explicit reasoning is needed, models with such a mode are available nearby in BotHub.
Support for function calling and strict JSON output is not noted in our data, so we cannot guarantee it. The easiest way to check is in practice: Extract is available via BotHub's unified OpenAI-compatible API, and you can switch to a model with confirmed function calling without rewriting the integration.
Images and documents—yes, these are confirmed input formats, and documents are processed natively without prior conversion to text. There is no note about audio and video in our data, so for transcribing recordings, it is better to choose a specialized model from the BotHub catalog.
We do not have data on the quality of performance with specific languages, so we will not promise anything specific—evaluate it on your task. In BotHub, it's simple: pay-as-you-go for tokens, no subscription, tokens do not expire, and access from Russia works without a VPN or foreign card.