Mistral OCR

[Dual-0]

Hi,

Mistral introduces Mistral OCR, a new Optical Character Recognition API that sets a new standard in document understanding by accurately comprehending media, text, tables, and equations from images and PDFs. It outperforms other OCR models in benchmarks, including overall accuracy, math, multilingual capabilities, and table recognition and is natively multilingual, supporting thousands of scripts and languages. Mistral OCR is processing up to 2000 pages per minute, and offers a "doc-as-prompt" feature for structured output. The API is available on la Plateforme.

Would be nice to use this feature in LibreChat to OCR PDFs.

[danny-avila]

Yea I'd like to implement this and then implement an open-source version of this. Uploading files as text is very timely since it's on the roadmap as an upcoming item.

[djuillard]

I am wondering whether this Mistral OCR thing is very different from what we can achieve today with well known parsing repo like docling.