Parse 5
Parse 5 converts document content into Markdown for downstream search and knowledge workflows. It extracts reading order, tables, lists, form fields and visual-element locations.
Capabilities and limits
Cohere documents stable parsing for nine languages. Parse does not return confidence scores, JSON output, or a hierarchy of headers and footers. The API is proprietary. The linked North Micro Vision checkpoint is a separately described open-weight model, not a documented downloadable copy of the Parse service.Parse documentation
When this model fits
Practical guidance. Parse is useful when many documents must become searchable. Review representative tables, columns, footnotes and forms before using the result as evidence. Preserve the original file and page reference alongside extracted content. If a downstream application needs a strict JSON record, add explicit field extraction and validation after parsing. A readable conversion alone does not establish that every cell or key-value pair is correct.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | parse-v5.0 |
| Context window | 8,192 tokens Published context limit for this API model. Parse documentation |
| Supported inputs | PDF, PPT and base64-encoded JPEGParse documentation |
| Model scale | 2.3 billion parametersParse documentation |
| Output | Markdown, HTML tables and bounding boxesParse documentation |
Related models
- Embed 5 Pro
Cohere · Embed
Quality-focused embedding model for text, images and mixed documents, sharing a vector space with Embed 5 Fast.
Launch date not confirmedAPI - Command A Vision
Cohere · Command
Vision-language model for documents, charts, tables and visual questions, with up to 20 images per request.
Launch date not confirmedAPI · Downloadable A 2.4-billion-parameter vision-language model for OCR, charts and document questions, with interleaved image and text input.
Launch date not confirmedDownloadable