|

Cohere Releases Parse 5 (parse-v5.0): A 2.3B Vision Language Model That Turns Enterprise Documents Into Markdown

ByRicardo

Cohere has launched Parse (parse-v5.0), a doc parsing mannequin aimed toward high-volume enterprise ingestion. It is a 2.3B-parameter imaginative and prescient language mannequin with an 8,192-token context window and a ~4.6GB footprint, constructed on Cohere Labs’ North-Micro-Vision-Instruct structure. Parse takes a PDF, PPT or JPEG web page as a base64-encoded information URI and returns Markdown containing textual content in studying order, tables rendered as HTML, lists, kind key-value pairs, picture descriptions and bounding field coordinates. There is not any separate OCR stage in entrance of it. Cohere costs the Parse API at $1.50 per 1,000 pages and positions the mannequin on price-performance somewhat than peak accuracy — a declare the corporate helps with a self-reported ParseBench rating of 79.2 that, as we element beneath, measures three of that benchmark’s 5 dimensions.

Is it deployable?

Yes, in manufacturing. Parse is mostly out there by means of the Cohere Parse API, Microsoft Foundry, AWS SageMaker, and single-tenant Model Vault. There is not any waitlist and no analysis license.

  • Which corporations: Mid-market groups that already run a RAG stack can begin on metered API calls with a free trial key. Large enterprises with residency or air-gap necessities go straight to Model Vault or personal deployment. Seed-stage startups can use it, however the economics solely begin to matter above roughly 100K pages a month.
  • Which industries: Cohere targets monetary providers, insurance coverage, healthcare and life sciences, public sector, telecom, vitality and manufacturing — the document-heavy verticals the place scanned varieties and dense tables are the norm.
  • Applications: RAG ingestion, clever doc processing, claims and bill pipelines, contract and submitting search, and giving doc context to brokers.

What is Parse?

Parse is a 2.3B-parameter imaginative and prescient language mannequin constructed on Cohere Labs’ North-Micro-Vision-Instruct structure, with an 8,192-token context window and a ~4.6GB footprint. It accepts PDF, PPT and JPEG pages as base64-encoded information URIs and returns Markdown containing doc textual content, lists, tables rendered as HTML, bounding field coordinates and picture descriptions.

There is not any separate OCR stage in entrance of it. The mannequin recovers textual content and studying order, tables, lists, varieties and key-value pairs, photos and captions, and the areas of web page boundaries and visible components in a single go. Nine languages are listed as secure — Arabic, English, French, German, Italian, Japanese, Korean, Portuguese and Spanish — with zero-shot assist elsewhere at decrease accuracy.

Two output modes matter in apply. The default returns a Markdown string per web page. Setting output_format="blocks" returns typed blocks, the place a desk block carries its HTML, its bounding field and an outline. That second mode is what makes citation-level traceability attainable.


The Benchmark

Cohere reviews a ParseBench rating of 79.2 for Parse, averaged throughout tables, content material faithfulness and semantic formatting, forward of Mistral OCR 4 (74.5), Azure Document Intelligence (74.3) and Databricks AI Parse (72.4).

ParseBench is a LlamaIndex benchmark of ~2,078 human-verified enterprise pages scored on 5 dimensions: tables, charts, content material faithfulness, semantic formatting and visible grounding. Cohere’s determine averages three of them and drops charts and visible grounding — the 2 dimensions the place most parsers collapse.

Against the public leaderboard, the identical distributors rating far decrease on the total five-dimension total: Mistral OCR 4 at 60.68, Databricks AI Parse at 60.68, Azure Document Intelligence (Layout) at 59.64. Azure’s three-dimension common works out to 74.3, which matches Cohere’s determine precisely and confirms the methodology. Cohere Parse is just not at the moment listed on that leaderboard, the place LlamaParse Agentic leads at 84.88.

So 79.2 is a vendor-reported subset rating, not a leaderboard place. It is an inexpensive declare to check by yourself paperwork.

What it Costs

Cohere costs the Parse API at $1.50 per 1,000 pages. On Model Vault, Parse 5 runs $4.00/hour or $2,500/month for a Medium occasion, and $7.00/hour or $4,300/month for XL.

The crossover no one publishes: at $0.0015 per web page, a single Medium occasion breaks even at roughly 1.67M pages per thirty days, and XL at roughly 2.87M. Below that, metered API calls are cheaper. Above it, devoted capability wins on value alone — earlier than any argument about information residency, which is normally the true purpose enterprises transfer to Vault.

Key Takeaways

  • Cohere shipped parse-v5.0, a 2.3B VLM that converts PDFs, slides and pictures into Markdown with HTML tables and bounding containers.
  • API pricing is $1.50 per 1,000 pages; Model Vault runs $2,500/month (Medium) or $4,300/month (XL).
  • Dedicated capability solely beats metered pricing above roughly 1.67M pages per thirty days.
  • The 79.2 ParseBench determine is vendor-reported throughout three of 5 dimensions and omits charts and visible grounding.

Check out the Technical details here and Try it on HF. Also, be happy to comply with us on Twitter and don’t overlook to affix our 150k+ML SubReddit and Subscribe to (*5*). Wait! are you on telegram? now you can join us on telegram as well.

Need to companion with us for selling your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar and many others.? Connect with us

The submit Cohere Releases Parse 5 (parse-v5.0): A 2.3B Vision Language Model That Turns Enterprise Documents Into Markdown appeared first on MarkTechPost.

Similar Posts