Contracts are the case where the cost of a mistake is known in advance, so the chain is built so that every step can be checked. First the document has to be read reliably along with its layout, then similar clauses and your approved wording have to be found, then differences compared and explained, and personal data has to be dealt with before anything leaves the building. The model here assists a lawyer rather than replacing one.
Contracts arrive as scans and PDFs where half the meaning sits in clause numbering, footnotes, annexes and tables. Document recognition models extract structure as well as text: clause, sub-clause, reference to an annex. Skip this step and the analysis drifts at the numbering level, so a reference to the penalty clause lands in the wrong section, which is unacceptable in legal work.
An early model that converts scientific PDFs into text with formulas. Now outdated and outperformed by almost all modern OCR models.
A model and toolkit for converting PDFs into clean text at scale, preserving reading order, tables and formulas. Built to process millions of pages.
A popular open tool for converting PDFs to Markdown with its own small model. MinerU2.5-Pro was improved through data alone, without growing in size. Languages on the card: Chinese and English.
Meaning-based search surfaces approved wording, earlier versions and previous contracts with the same counterparty. Some models search the document page directly, layout included, which suits scans. Without this step a lawyer reads every contract from scratch, and the company renegotiates points it already settled two years ago.
Searches PDFs and scans as images: pages do not need to be OCR'd first, the model finds the right one for a question directly, including tables and charts. Trained on English.
A model for meaning-based search in about a hundred languages. The core of RAG: the bot finds the right part of a document before answering.
Strong multilingual rerankers; m0 also ranks pages as images (scans, slides). The latest versions are open for non-commercial use only.
A language model matches the incoming text against your approved version and lists the differences in plain words: payment terms, liability, termination, jurisdiction. It flags what looks off while the lawyer decides, so every remark must point at a specific clause. Without this part a first pass takes hours, and small edits by a counterparty get noticed after signing.
Language models for legal texts, fine-tuned on US and European legal corpora (based on Mistral and Mixtral). English only.
A family of language models with strong Russian language support, from small versions for a laptop to a flagship on par with commercial APIs.
DeepSeek's flagship line: from the first 7B/67B to V4-Pro with 1.6 trillion parameters. Closed-model quality under an open MIT license; V4-Flash-Vision-Exp and V4.1-Flash understand images, context up to 1M tokens.
A dedicated model finds personal data and identifiers in the text and replaces them with placeholders before the document goes to an outside service or a shared archive. It works narrowly and fast and rewrites nothing else. Without this part, any external processing of contracts becomes a transfer of people data that you will have to account for, and that is no longer a technical question.
Small GLiNER-based models for finding personal data: passports, phone numbers, accounts, addresses. Data types are set in words. Russian is not officially supported.
A popular detector of 17 types of personal data in six European languages. No Russian and a non-commercial license: suitable for trials and research.
Finds and hides personal data: names, addresses, phone numbers, emails, account numbers, passwords. Runs even in the browser. Trained mostly on English.
Here more than anywhere, remember that a model is wrong confidently: a missed clause looks exactly as calm as a caught one. So the chain is deployed as a first pass and a risk highlighter rather than as approval, and every remark is verified by a lawyer against the text. Keep a log of who accepted what, or nobody will be able to trace how a wording ended up in the signed version. Legal questions, personal data and permitted processing are described here in general terms only, with a lawyer reviewing the specific case.
An open model can run on your own server: data stays in-house, there is no per-request fee, and the model can be fine-tuned on your documents.
