Cohere Releases Parse 5 (parse-v5.0): A 2.3B Vision Language Model That Turns Enterprise Documents Into Markdown

Details

Cohere has launched Parse 5 (parse-v5.0), a 2.3B-parameter vision language model built on its North-Micro-Vision-Instruct architecture, designed for high-volume enterprise document ingestion. The model accepts PDF, PPT, and JPEG pages as base64-encoded URIs and returns structured Markdown—text in reading order, HTML-rendered tables, lists, form key-value pairs, image descriptions, and bounding box coordinates—all in a single inference pass without a preceding OCR stage. Parse runs with an 8,192-token context window and a ~4.6GB footprint. Cohere prices it at $1.50 per 1,000 pages, betting on price-performance over peak accuracy. Its self-reported ParseBench score of 79.2 is worth noting with a caveat: it covers only three of the benchmark's five dimensions. Nine languages carry stable accuracy guarantees (Arabic, English, French, German, Italian, Japanese, Korean, Portuguese, Spanish), with zero-shot support for others. Two output modes matter for practitioners: the default returns per-page Markdown strings, while output_format=blocks returns typed blocks with bounding boxes and descriptions—enabling citation-level traceability in RAG pipelines. The model is generally available today with no waitlist across Cohere API, Microsoft Azure AI Foundry, AWS SageMaker, and Model Vault for enterprises with air-gap or data-residency requirements. Cohere targets financial services, insurance, healthcare, public sector, telecom, and manufacturing—verticals where scanned forms and dense tables dominate ingestion pipelines. The economics become compelling above roughly 100K pages per month.