Tasks and inputs
Cohere Parse is listed for Document question answering, Image-to-text, Document OCR. It accepts image, pdf and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Cohere Parse (parse-v5.0) is a 2.3B vision-language document parser that turns PDFs, slides and scans into Markdown for RAG, at $1.50 per 1,000 pages.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: $1.50 per 1,000 pages on the Cohere API; Model Vault pricing is per deployment. Confirm with Cohere.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Ingestion is the unglamorous half of retrieval, and it is usually where quality leaks out. Cohere Parse is the company's answer to that: a compact vision-language model, parse-v5.0, that reads PDFs, PowerPoint decks and JPEG scans and returns Markdown a retriever can actually use.
It launched on 27 August 2026 and slots in beside Cohere's Embed and Rerank models in the same pipeline.
Send a Base64-encoded file to the /v2/parse endpoint with output_format set to markdown. What comes back:
Under the hood it is about 2.3 billion parameters (roughly 4.6 GB), built on the same architecture as North Micro Vision Instruct. The context length is 8,192 tokens.
| Spec | Value |
|---|---|
| API ID | parse-v5.0 |
| Launched | 27 August 2026 |
| Inputs | PDF, PPT, JPEG (Base64) |
| Stable languages | Arabic, English, French, German, Japanese, Korean, Italian, Portuguese, Spanish |
| API price | $1.50 per 1,000 pages |
| Throughput | 4.5 pages/s on one 8xH100 node |
| Platforms | Cohere API, Model Vault, Microsoft Foundry, AWS SageMaker |
The gaps are specific, and Cohere is upfront about them. There are no confidence scores. Headers, footers and font hierarchy are not detected. There is no structured JSON mode, so if your downstream system wants typed fields you still need an extraction step after parsing. Bounding boxes cover tables and images but not ordinary text lines, and chart data extraction is promised for a later version.
Languages outside the nine listed run zero-shot, with lower accuracy likely.
Cohere ran ParseBench itself and reports 79.2 overall, with 87.0 on tables, 86.6 on content faithfulness and a weaker 64.0 on semantic formatting. It places that above Mistral OCR 4 at 74.5 and above several managed parsing services; Mistral has since moved on to OCR 4.1, which was not in the comparison. Layout and chart categories were left out of the comparison, which flatters any model without chart support. Use the numbers as a shortlist filter, not a verdict.
At the API rate, a million pages is $1,500. Cohere argues that a dedicated Model Vault deployment costs 23% less than the API at half GPU utilisation and 61% less at full utilisation. That calculation only works if you can keep the hardware busy. Treat both figures as indicative and confirm them with Cohere for your own contract.
Pull 200 pages that broke your current parser: merged-cell tables, two-column layouts, scanned forms. Parse them, diff the Markdown against hand-checked output, then run your actual retrieval evaluation on top. Parser scores matter less than whether the right chunk comes back. Compare against other options in document OCR, and see the rest of the stack on the Cohere provider page.
Compare the model on the work you need to complete.
Cohere Parse is listed for Document question answering, Image-to-text, Document OCR. It accepts image, pdf and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 8.19K context. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
OpenBMB
MiniCPM-OCR 1.0 is OpenBMB's open 1.3B document-parsing model for text, tables, formulas and charts, with a free API and on-device apps.
OpenBMB
MiniCPM-OCR Jev is OpenBMB's open 2B judge that scores whether an OCR result for a text, formula or table region is correct, from 0 to 1.
Tencent
Youtu-Parsing-Omni is Tencent Youtu's 5B open parser that turns documents, charts, images, audio and video into one structured JSON format.
lightonai
LightOnOCR-3-0.8B is LightOn's smallest open document model: page images in, markdown out, with optional bounding boxes and chart tables.
Sarvam
Sarvam Vision 2.1 is Sarvam's 3B document-intelligence VLM for OCR, tables, forms and Indic handwriting in 22 Indian languages plus English.
NVIDIA
Nemotron 3.5 Parse is NVIDIA's 0.631B open document parser that turns page images into Markdown, HTML or JSON with tables, charts and reading order.