Cohere releases Parse v5.0 vision parsing model

Cohere has released Parse v5.0, a high-throughput vision language model designed to process large volumes of enterprise documents into structured Markdown data. The model supports nine major languages and is available via the Cohere API, Model Vault, Microsoft Foundry, and AWS SageMaker.

• On ParseBench, Parse scores 79.2, outperforming Mistral OCR 4 (74.5), Databricks AI Parse (72.4), LlamaParse (78.3), and hyperscaler solutions like AWS Textract and Google Document AI by over 20 points.
• It processes 4.5 pages per second on a single GPU, achieving approximately 1.4x the throughput of RedNote's dots.mocr and 2.2x that of Chandra OCR 2.
• Deploying through Model Vault offers significant cost savings, reducing inference costs by up to 61% compared to the standard API at full utilization.

The model is positioned as a cost-effective solution for document indexing, RAG, and agentic retrieval, preserving document structure for downstream automation.