Cohere Parse 5 Undercuts Rivals on Price Despite Trailing on Accuracy

Quick Facts

• Cohere released Parse 5 on Aug. 27, 2026, priced at $1.50 per 1,000 pages through its API.
• The model scores 79.2 on three benchmark dimensions, trailing GPT-5.5 (84.4) and Gemini 3.5 Flash (81.8).
• Parse 5 processes 4.5 pages per second and is available on Microsoft Foundry and AWS SageMaker.

Cohere launched Parse 5 on Aug. 27, a vision language model designed to convert enterprise documents into structured Markdown. The company is not claiming the top benchmark score. It is claiming the best cost-per-page for enterprise scale.

The model is priced at $1.50 per 1,000 pages through the Cohere API. That undercuts common hyperscaler alternatives priced at $10 per 1,000 pages by a wide margin. For a large enterprise accounts payable workflow processing 13 million document pages per month, switching to Parse 5 via Model Vault could save roughly $1.47 million annually compared to a $10-per-1,000-page service.

Parse 5 is a 2.3-billion-parameter vision language model with an 8,192-token context window and a 4.6-gigabyte footprint. It accepts PDFs, PowerPoint files, and JPEGs as base64-encoded inputs and returns Markdown with tables rendered as HTML, form key-value pairs, image descriptions, and bounding box coordinates. There is no separate OCR stage.

On throughput, Cohere says Parse 5 processes 36 pages per second on an 8-GPU H100 node, about 1.4 times faster than RedNote’s dots.mocr and 2.2 times faster than Chandra OCR 2 on the same hardware.

The Benchmark Question

Cohere reports a composite score of 79.2 across three dimensions: tables, content faithfulness, and semantic formatting. That places Parse 5 behind GPT-5.5 at 84.4, Opus 4.8 at 84.3, and Gemini 3.5 Flash at 81.8. It beats LlamaParse’s Cost Effective tier at 78.3, Mistral OCR 4 at 74.5, Databricks AI Parse at 72.4, and Azure Document Intelligence at 69.3.

The benchmark itself comes with a caveat. ParseBench was created by LlamaIndex, which also sells a competing product called LlamaParse. The full benchmark scores five dimensions; Cohere’s reported figure drops two of them: layout and chart extraction. Cohere says both are outside the current product scope. On the full five-dimension leaderboard, competitors score significantly lower: Mistral OCR 4 and Databricks AI Parse each land at 60.68, and Azure Document Intelligence at 59.64.

Nils Reimers, VP of AI Search at Cohere, told VentureBeat: “Document parsing isn’t solved because the hard part isn’t reading text, it’s preserving structure and meaning. Enterprise documents mix tables, diagrams, charts, and formatting that change the interpretation of the data.”

Deployment and Pricing Structure

Parse 5 is available now through the Cohere API, Model Vault, Microsoft Foundry, and AWS SageMaker. Model Vault, Cohere’s single-tenant managed inference platform, runs $4.00 per hour or $2,500 per month for a Medium instance and $7.00 per hour or $4,300 per month for an XL instance.

The economics favor dedicated capacity at scale. A Medium instance breaks even against metered API calls at roughly 1.67 million pages per month. Above that volume, the fixed monthly rate wins on price before any data residency argument enters the picture. For regulated industries, the data residency point may matter more: Parse deployed in Microsoft Foundry stays within the Azure tenant boundary and inherits existing network and identity controls.

Where Cohere Is Betting

Cohere targets financial services, insurance, healthcare, and public sector customers where scanned forms and dense tables are standard. Use cases include RAG pipeline ingestion, claims and invoice processing, and feeding document context to AI agents.

The architecture supports a cost-reduction pattern some engineering teams are already using: run Parse 5 first to extract clean Markdown, then pass that output to a lighter reasoning model. That split cuts the total cost of a document pipeline and lets teams swap reasoning models without retraining the document reader.

Parse 5 also fits a broader Cohere strategy. The company has been moving from selling foundation models to positioning itself as an enterprise data platform. A model that sits at the start of every document workflow is a defensible position, regardless of where it lands on any single benchmark.

Read more: Cohere Parse 5 loses the benchmark on points. It wins on cost per page.