Publication date: 28 August 2026
Cohere has released Parse, model ID parse-v5.0, a 2.3-billion-parameter vision language model for turning complex enterprise documents into structured Markdown. The API costs $1.50 per 1,000 pages and is also available through Microsoft Foundry, AWS SageMaker and Cohere Model Vault.
The release targets a practical bottleneck in enterprise AI: converting tables, forms, images and mixed-layout documents into reliable input for retrieval and agent workflows.
What does Cohere Parse extract?
Parse is designed for document understanding rather than general chat. It extracts text in reading order, tables, lists, forms, images, captions, page boundaries and selected visual locations. Cohere lists an 8,192-token context window and a model size of about 4.6 GB.
The documented input types include PDF, PowerPoint and JPEG, while the main output is Markdown. The API can also return structured table content and visual metadata for downstream processing.
| Specification | Cohere Parse |
|---|---|
| Model ID | parse-v5.0 |
| Parameters | 2.3B |
| Context length | 8,192 |
| Model size | ~4.6 GB |
| API price | $1.50 per 1,000 pages |
| Main output | Markdown |
How does Cohere Parse compare on document parsing?
Cohere reports a ParseBench average score of 79.2 across table extraction, content faithfulness and semantic formatting. In the same vendor-run evaluation, Mistral OCR 4 scored 74.5 and LlamaParse Cost Effective scored 78.3. Gemini 3.5 Flash scored 81.8.
These are Cohere’s evaluation results, not an independent benchmark. Cohere says it re-scored competitors against updated ParseBench rules. The company also excludes chart extraction and part of layout detection from the headline average because those capabilities fall outside the current product scope.
| Model | ParseBench average | Position |
|---|---|---|
| GPT-5.5 | 84.4 | General-purpose frontier model |
| Gemini 3.5 Flash | 81.8 | General-purpose frontier model |
| Cohere Parse | 79.2 | Specialist document parser |
| LlamaParse Cost Effective | 78.3 | Specialist parser |
| Mistral OCR 4 | 74.5 | Specialist parser |
Cohere Parse pricing and deployment
The public API price of $1.50 per 1,000 pages makes the model easy to budget for high-volume ingestion. Cohere also offers Parse in Model Vault for single-tenant inference. The company says Model Vault can reduce inference cost further at sustained utilization.
A procurement or knowledge-platform team should still check total pipeline cost. Document conversion quality affects later embedding, retrieval and agent performance. A cheaper parser can become expensive if it creates malformed tables or drops context that causes downstream retries.
What this means
Worth testing now for teams processing large volumes of invoices, reports, forms or scientific documents. The model is small enough to be operationally interesting, and the per-page API price is clear enough to run a meaningful pilot without a complex pricing model.
Start with your hardest internal documents, especially multi-column pages, dense tables and scanned forms. Measure extraction accuracy and downstream retrieval quality together. Cohere’s own benchmark notes that chart data extraction is not yet part of the current product scope, so chart-heavy workloads need separate testing.
For more information, visit the official announcement of Cohere Parse on the Cohere website.
