Cohere Parse 5 pricing targets volume, trails rivals in reported benchmark

Cohere Parse 5 pricing is $1.50 per 1,000 pages, or $0.0015 a page, as the company introduces a document-parsing model aimed at high-volume enterprise ingestion. The model trails GPT-5.5, Opus 4.8 and Gemini 3.5 Flash in Cohere’s reported three-dimension ParseBench results, putting the launch’s case on operating cost rather than peak score.

Cohere announced Parse on Aug. 27, describing it as a 2.3-billion-parameter vision-language model that converts PDF, PowerPoint and JPEG pages into structured Markdown. Cohere says the model processes a page image in one model pass rather than using a separate OCR step, returning text in reading order and HTML tables. It is generally available through Cohere’s API, Model Vault, Microsoft Foundry and AWS SageMaker.

The listed rate works out to $1,500 for one million pages. That establishes Parse 5’s own API price, but the available evidence does not provide like-for-like per-page pricing for competing models. Cohere’s claim that it has the strongest price-performance profile therefore remains a vendor claim, rather than a verified market-wide cost comparison.

How does Cohere Parse 5 compare on price and accuracy?

In Cohere’s reported ParseBench table, Parse 5 scored 79.2 across tables, content faithfulness and semantic formatting. GPT-5.5 scored 84.4, Opus 4.8 scored 84.3 and Gemini 3.5 Flash scored 81.8. The same comparison put 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.

Those results should not be read as a complete ranking. ParseBench evaluates tools against human-verified enterprise pages, but Cohere’s displayed comparison left out the Layout and Chart dimensions. Cohere attributes the omission to product differences: Parse 5 provides reading-order Markdown instead of bounding boxes for each text element, and it describes charts rather than extracting their underlying data. The company says chart-data extraction is planned for a future version.

That distinction is operational rather than cosmetic. A parsing system turns documents into material that retrieval and agent systems can use. If a workflow depends on precise chart values, element-level coordinates or complex layouts, buyers need to test whether Parse 5’s output format preserves the required information.

What should enterprise buyers test?

Cohere’s Nils Reimers said the company modeled a financial-services workflow handling 750 million documents a year and estimated that Parse 5 would reduce costs by more than 98% versus GPT-5.5. The figure is Cohere’s estimate for one modeled workflow, not an audited deployment.

Stephanie Walter, practice leader for AI Stack at HyperFRAME Research, told VentureBeat that companies should evaluate parsers against their most difficult documents and measure downstream retrieval and task results, rather than extraction appearance alone. That approach aligns with a broader framework for evaluating AI models for the work they will actually do: test representative inputs, define acceptable failures and compare total workflow cost under the same conditions.

• Use documents containing tables, forms, charts and irregular reading order.
• Check whether output retains the provenance and structure required downstream.
• Compare API, deployment and model costs against the same page mix and accuracy threshold.

For teams processing millions of pages, Cohere’s listed rate is concrete. Whether it produces the lowest usable cost depends on the documents, required output and error tolerance, none of which a three-dimension benchmark or a published list price settles on its own.

This story draws on original reporting from VentureBeat.