Top AI Product

Cohere shipped Parse 5 (parse-v5.0) on August 27: a 2.3B-parameter vision language model that turns PDFs, PPTs, and scanned images into structured Markdown. Self-reported ParseBench score is 79.2 — behind GPT-5.5 (84.4) and Opus 4.8 (84.3). Cohere isn’t hiding that. The whole pitch is cost per page, and when you’re feeding millions of contracts into a RAG pipeline, $1.50 per 1,000 pages beats five extra benchmark points.

What it actually does

Text comes back in reading order, tables as HTML, forms as key-value pairs, images as descriptions with bounding box coordinates. A blocks mode returns typed elements, so an agent can cite the exact table it pulled a number from. The model is ~4.6GB with an 8,192-token context — small enough to run private deployments without a GPU cluster.

The API

Send base64-encoded pages to the Parse API, get Markdown back. It’s also live on Azure AI Foundry and AWS SageMaker, with Model Vault private deployment from $2,500/month. Typical use cases: insurance claims, invoice pipelines, contract search — the dirtiest, most unavoidable step in every enterprise RAG stack. Document parsing rarely trends; a small model attacking it on price is why this one did.

You Might Also Like

• Cursor Agent Swarms hit 1000 Commits per Second and Dropped Worker Cost From 9373 to 411
• Starnus Just hit 1 on Product Hunt and Yeah its Worth the Hype
• Lovon Just Topped Product Hunt on Valentines day and its not a Dating app
• Zenmux Just hit 1 on Product Hunt Heres why Everyones Paying Attention
• Mercury 2 Just hit 1000 Tokens per Second and its not Even Using Transformers