Cohere launches Parse, a document-conversion model for enterprises

Cohere launched Parse, a vision-language model built to convert multilingual, multimodal enterprise documents — reports, forms, scanned files — into structured Markdown for use in search, retrieval and agentic workflows. The model handles documents and images across nine major world languages, with additional file formats such as spreadsheets and web pages supported when used through Cohere’s Compass product.

Cohere reported Parse scoring 79.2 on its own ParseBench evaluation, against 53.3 for Amazon’s Textract and 57.3 for Google’s Document AI — figures that are the company’s own comparison rather than independently reproduced — with the largest gaps on semantic formatting (64.0 versus 2.8 and 33.0) and on preserving table structure. The model is priced at $1.50 per 1,000 pages through Cohere’s API, with the company saying costs can fall by as much as 61% at full utilisation for customers running it through its Model Vault deployment option.

Parse is available through Cohere’s own API, Model Vault, Microsoft Foundry and AWS SageMaker, placing it in direct competition with the document-processing services built into those same cloud platforms. The launch continued Cohere’s focus on enterprise infrastructure — search, retrieval and now document conversion — rather than competing on general-purpose chat performance against the largest frontier labs.