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

Kwon Crash

Published Aug 27, 2026, 9:53 PM UTC

Source: AISource
- Cohere dropped Parse 5 — a 2.3B vision language model that eats PDFs, slides and JPEGs and spits out Markdown with HTML tables and bounding boxes. No separate OCR stage, just one pass. API runs $1.50 per 1,000 pages, or you can grab a Model Vault instance from $2,500/month. The crossover nobody advertises: dedicated capacity only beats metered pricing above roughly 1.67M pages per month. Below that, pay per page and keep your stack-eye on the bill. Now the spicy part — Cohere boasts a ParseBench score of 79.2, ahead of Mistral OCR 4, Azure Document Intelligence and Databricks AI Parse. Except that number averages three of the benchmark's five dimensions and conveniently drops charts and visual grounding, the