Cohere launches Parse 5: a vision language model for converting enterprise documents into markdown

Cohere has launched Parse 5 (parse-v5.0), a vision language model designed to convert enterprise documents into structured Markdown. The model, which has 2.3 billion parameters, operates on Cohere Labs’ North-Micro-Vision-Instruct architecture and features an 8,192-token context window and a ~4.6GB footprint. It accepts inputs such as PDFs, PPTs, and JPEG pages encoded as base64 data URIs and outputs Markdown containing text in reading order, tables rendered as HTML, lists, form key-value pairs, image descriptions, and bounding box coordinates. Unlike some competitors, it does not require a separate OCR stage, processing text, tables, lists, forms, and visual elements in a single pass. The model supports nine languages, including Arabic, English, French, German, Italian, Japanese, Korean, Portuguese, and Spanish, with zero-shot support for other languages at lower accuracy. Parse offers two output modes: the default returns Markdown per page, while the 'blocks' mode provides typed blocks with HTML, bounding boxes, and descriptions, enabling citation-level traceability. Cohere reports a ParseBench score of 79.2, which averages performance across tables, content faithfulness, and semantic formatting, outperforming Mistral OCR 4 (74.5), Azure Document Intelligence (74.3), and Databricks AI Parse (72.4). The ParseBench benchmark evaluates five dimensions, including tables, charts, content faithfulness, semantic formatting, and visual grounding, with Cohere’s score excluding charts and visual grounding, where most parsers struggle. Cohere prices the Parse API at $1.50 per 1,000 pages. On Model Vault, Parse 5 runs at $4.00/hour or $2,500/month for a Medium instance, and $7.00/hour or $4,300/month for an XL instance. The break-even point for a Medium instance is approximately 1.67 million pages per month, while an XL instance reaches 2.87 million pages. Below these thresholds, metered API calls are more cost-effective, while dedicated capacity becomes preferable above them, especially considering data residency concerns. The Parse model is currently available through the Cohere Parse API, Microsoft Foundry, AWS SageMaker, and single-tenant Model Vault, with no waitlist or research license required. Cohere also highlights technical details and invites users to try the model on Hugging Face. The post originally appeared on MarkTechPost.